The Business Engineer

Preview the book & course module here!

Gain Full Access to The Research & Atlas

If you’re already a paid yearly member, simply reply to this email, and we’ll send it your way.

The Personal AI Agents Map

Gennaro Cuofano — Sun, 04 Oct 2026 12:59:37 GMT

Having spent over a decade in the AI space, I still remember how, back in 2017, we were trying to build voice agents, then called voice assistants, on platforms like Amazon Alexa and Google Assistant. Seeing where we are today makes two things clear to me: the possibilities are enormous, and we are still incredibly early.

OpenClaw was the real aha moment for what personal agents could do, both for business users and consumers. In less than a year since its inception, the landscape has evolved dramatically. Things are moving incredibly fast, especially for business users, while consumer interfaces that initially looked like “OpenClaw copycats” are beginning to specialize and develop identities of their own.

For the vast majority of consumers, an agentic experience is still something they have never encountered. These new personal agents will be their first point of contact with AI that can actually act on their behalf.

As a business user working at the edge of the AI industry, interfaces like Instinct or Muse didn’t impress me technically in the same way that Claude Code, OpenClaw or Grok Bot did. Those harnesses offered a glimpse of the full capabilities of agentic AI. But my perspective isn’t necessarily the one that determines what succeeds with consumers.

Consumer AI is a different game. It sits at the intersection of interface design, monetization that can sustain free access at scale, and brand recognition. The most technically impressive harness won’t automatically become the interface millions of people use every day.

That is what makes this moment so interesting: the personal agent landscape is finally emerging in full swing, and we are beginning to see how these capabilities might reach a mass audience.

The personal agent is becoming a new layer between intent and fulfilment. The fight is no longer only over which model you use. It is over who receives the first request, who earns the right to act, and who gets paid when the work is done.

A user sends a voice note: “I need to be in New York tonight.” Nobody opens an airline app, compares hotels, checks Uber, copies the itinerary into a calendar or types a card number. The agent already knows the user’s location, preferred airline, seat, payment method and schedule. It books the flight, reserves the hotel, orders the car and updates the calendar. Earlier that day, the same agent may have called first because a document needed signing before 3 p.m.

That is how Noah Shinn describes Instinct. The product has no conventional app. It has a phone number, an email address and a computer. You deal with it much the way you would deal with another person. down

For the first few years of consumer generative AI, the chat window was the product. You opened ChatGPT, Claude, Gemini or another application, asked for something, received an answer, then completed the workflow yourself. The assistant sat beside the work.

The new personal agents are trying to sit in front of the work.

They remember. They watch. They decide when something requires attention. They browse, call tools, spend money, coordinate with other people and increasingly continue working after the user has gone away.

By September 2026, Meta had launched Muse, Apple had put Siri AI into beta, Anthropic had folded Cowork into Claude, Rabbit had released OS3, Microsoft had announced Autopilot and OpenAI had launched Dots.

Google had already been running Gemini Spark, Amazon had pushed its shopping agent deeper into Alexa, and Chinese internet groups were reported to be testing their own versions. Instinct announced a $1 billion financing round at a $10 billion valuation in the same period.

The products differ enormously underneath, but they are converging on the same architectural idea:

The personal agent is becoming a new layer in the stack: the delegation layer.

Search engines spent two decades owning the moment when the user asked, Where can I find this? Marketplaces fought to own the moment when the user asked, Where should I buy it? The personal agent wants to own an earlier moment:

“Here is what I want. Take care of it.”

That position is potentially more valuable than the search box or the checkout page because it sits before both. The agent can decide where to search, which service to use, which merchant to trust, whether a transaction is worth making and whether the user needs to see the intermediate steps at all.

The category becomes much easier to understand through six questions: Where does the agent run? How do you reach it? Who starts the work? Who pays? How does it touch other services? Who else can it talk to?

If you are a paid member, reply to this to request access.

Also watch the full video lecture below!

Inference Engineering: The Tokenomics of AI

Gennaro Cuofano — Sat, 03 Oct 2026 11:59:12 GMT

For almost four years, the AI Supercycle was dominated by one question: how do we build more capable models?

First came pretraining. Scale the Transformer with more data, more parameters, and more compute, and increasingly powerful capabilities emerged. Then came post-training. Instruction tuning, RLHF, reinforcement learning, verifiable rewards, and eventually reasoning techniques turned those pretrained models into systems that could follow instructions, solve harder problems, and interact with people in useful ways.

That was the first phase of the cycle. Most of the industry’s attention, capital, and technical prestige went into creating intelligence.

We are now moving into a different phase: the inference phase.

Training does not become less important. Frontier models will continue to require enormous amounts of compute. But training happens in large, concentrated runs. Inference happens every time the model is used. Every question, search, document review, coding task, reasoning step, tool call, and agent action requires the machine to run again. Google reported processing more than 3.2 quadrillion tokens per month as read in 2026, around seven times the level of a year earlier. At the same time, reasoning models are producing longer outputs, context windows have expanded dramatically, and agents can invoke models dozens of times to complete a single piece of work.

My expectation is that over the next three to five years, enterprise inference becomes at least as important economically as the pretraining layer, and potentially much larger. The reason is straightforward: once intelligence becomes useful enough, the economic problem shifts from producing the model to operating it continuously across millions of workflows.

And this is where something important changes.

For years, inference looked like plumbing. You picked a model, called an API, paid for the tokens, and focused on the application above it. But at enterprise scale, that abstraction starts to break down. The price of the model is only one part of the cost. The economics increasingly depend on how the request is served.

A model request actually contains two very different jobs.

Prefill is reading. The model processes the prompt, the instructions, the documents, the conversation history, and everything else in the context. Those input tokens can largely be processed in parallel, which makes prefill relatively compute-intensive and strongly influences how long the user waits before the first token appears.

Decode is writing. Once the model starts answering, it generates one token after another because each new token depends on what came before. That makes generation far more sensitive to memory bandwidth and to how many conversations share the hardware at the same time.

That distinction sounds technical, but it determines the economics of enterprise AI.

A system that reads a 100-page contract and returns one classification has a very different cost structure from a coding agent generating thousands of tokens from a short instruction. An agent that repeatedly rereads 50,000 tokens of context over thirty steps has a different profile again. The same model, at the same token price, can produce radically different costs depending on what it is asked to read, generate, repeat, and remember.

This is why inference engineering is becoming a discipline of its own.

It is about batching requests so expensive hardware stays productive without making users wait too long. It is about managing the KV cache, the working memory of each conversation, so long contexts do not consume the entire machine. It is about reusing prefixes instead of repeatedly paying to read the same instructions and documents. It is about deciding which work needs real-time capacity, which can run in batch, when dedicated infrastructure makes sense, and when renting tokens remains cheaper.

Above all, it is about changing the unit of optimization.

The objective is not the cheapest token.

It is the lowest cost per accepted outcome at the latency the workload actually requires.

That becomes even more important as reasoning and agents spread. Reasoning models can consume more computation to solve harder problems. Agents multiply inference calls and repeatedly reread growing contexts. The cost per token may keep collapsing while the number of tokens required to perform useful work rises much faster. That is how intelligence can become dramatically cheaper while the total inference market becomes dramatically larger.

So this piece goes beneath the API and into the machinery that increasingly determines the real economics of AI: prefill and decode, working memory, batching, caching, serving modes, latency, utilization, hardware, and the software engines connecting them all.

The concepts are technical, but the logic is surprisingly simple.

Serve to the target. Keep the cache warm. Pay for busy, not idle.

That is the operating system of the inference phase.

Preview the book & course module here!

For the last three years, I’ve been rebuilding the Business Engineer’s curriculum from the ground up. That curriculum has now become the foundation of a new discipline, with the entire series taking shape around it.

If you’re already a paid member, simply reply to this email, and we’ll send it your way.

Preview the book & course module here!

The Full Enterprise AI Stack

Gennaro Cuofano — Wed, 30 Sep 2026 05:00:11 GMT

This is a resource that any CTO should read. Why?

For almost four years, much of the AI conversation has centered on how intelligence gets built. Pretraining gave models broad capabilities by learning from enormous datasets. Post-training shaped those capabilities into more useful behavior. Reinforcement learning from human feedback, or RLHF, helped turn language models into assistants that could follow instructions and interact with people naturally. These were not separate technologies replacing one another, but successive layers of improvement around the model.

We have now entered what I would call the inference phase of the AI Supercycle. Not because training has stopped mattering, but because the economic question is expanding from How do we build a capable model? to How do we operate that capability repeatedly, across millions of tasks, at a cost the customer can justify? Reasoning models make this transition particularly important: additional computation during inference can itself improve performance. The model’s capabilities depend not only on how it was trained, but also on how much work the system performs when answering.

My expectation is that over the next three to five years, production inference could become a larger spending pool than pretraining alone, with enterprise deployment an important driver. Evidence already points in that direction within a defined market: Gartner forecasts inference to account for 55% of AI-optimized cloud infrastructure spending in 2026 and 59% in 2027, exceeding training spending in that segment. This isn't an industry-wide accounting of every AI investment, but it shows why inference deserves to move from a technical consideration to the center of enterprise strategy (see Gartner).

The mechanism is straightforward. Training costs recur as models are developed and improved. Inference costs recur whenever those models are used. And an agentic workflow can require much more than a single answer. Imagine a system investigating a customer issue: it reads the account history, checks a contract, chooses a tool, processes the result, revises its plan, and prepares an action. Each model call adds work. Multiply that pattern across departments, customers, and continuous operations, and serving intelligence becomes a substantial production system.

But inference is not one uniform workload, and that is where the nuances become fundamental. In a typical autoregressive language model, two phases do different jobs:

Prefill processes the input. The model works through the prompt, retrieved documents, instructions, and other context, building intermediate attention state commonly held in a key-value, or KV, cache. Much of this processing can happen in parallel. For sufficiently long inputs, available computation is often the main constraint, and prefill contributes substantially to the wait before generation begins.
Decode generates the output. The model produces subsequent tokens sequentially, using the context and previously generated tokens. Especially at smaller batch sizes, performance is often constrained by how quickly weights and cached state can move through memory rather than by arithmetic capacity alone. This affects how quickly the response progresses after generation begins.

That distinction has direct business consequences. A system reading a large contract pack to return a short classification has a different workload profile from one generating a long software change from a short instruction. An agent repeatedly using the same context introduces another consideration: whether the serving system can reuse previous computation instead of paying to process the same material again. The same model can have very different economics depending on what the application asks it to read, generate, and repeat.

This is why serving architecture matters. Prefill and decode can sometimes benefit from separate resource pools, different scheduling, and different capacity allocations. But separation isn't automatically better: transferring cached state adds overhead, and short or heavily cached requests may be more efficient without it. The right architecture follows the workload, not a universal rule about which chip, cloud, or model is best.

For an enterprise, this turns apparently technical choices into operating decisions. How much context should each request receive? Which repeated inputs can be cached? Which tasks can wait for batch processing? Which require immediate responses? Where does a more capable model reduce retries and human review enough to justify its price? The objective is not simply cheaper tokens. It is lower cost per accepted outcome, with quality, latency, and reliability held to the required standard.

That is also why this phase requires a map of the whole stack. Inference depends on the physical infrastructure underneath the model, but useful enterprise work depends equally on the context, routing, tools, permissions, memory, and evaluation around it. The company must decide not only where intelligence runs, but where the knowledge and authority that make it useful accumulate. Those are the decisions that determine whether it can change providers without rebuilding its operating capability.

The first phase made us ask how powerful the model could become. The inference phase forces us to understand the system that turns that capability into recurring work. This piece maps that system, from power and silicon to serving, agents, and enterprise control, and asks which parts should be rented and which must remain under the company’s control.

Rent the layers. Own the seams. Price the exit.

Preview the book & course module here!

If you’re already a paid member, simply reply to this email, and we’ll send it your way.

Preview the book & course module here!

The Agentic Web Map

Gennaro Cuofano — Mon, 28 Sep 2026 04:59:45 GMT

I have always been a student of history, and history has a strange property: the deeper you dig, the less clean its lessons become. You start looking for simple explanations and instead find contingencies, accidents, incentives, technologies, personalities, and decisions interacting with each other. Yet that is precisely why history matters. It cannot tell us what will happen next, but it can prevent us from repeatedly walking into the same traps.

Working in technology added another intellectual habit to that historical one: Silicon Valley’s obsession with first-principles thinking. It is a powerful idea, but also one surrounded by mythology. We all like to imagine ourselves stripping a problem down to fundamental truths and rebuilding it from scratch. Reality is less flattering. Almost all of us borrow from each other. Most innovation is recombination. Only a tiny number of people in any field genuinely redefine its first principles.

For the business person, that is not necessarily a disadvantage. In fact, one of the most valuable skills a generalist can develop is the ability to move across disciplines without becoming overly loyal to any of them. Specialization creates extraordinary depth, but it also creates boundaries. Finance speaks to finance. Computer science to computer science. Economics to economics. Organization theory to organization theory. The operator does not have that luxury because reality does not respect departments.

This is also why the best entrepreneurs often remind me more of scientists than of academics at their most specialized. The scientist has a hypothesis and eventually has to surrender it to reality. The entrepreneur does too. The business either works or it does not. Customers either care or they do not. The unit economics eventually show up. Reality has very little respect for ideological purity.

And many of the largest technological breakthroughs come from precisely this violation of disciplinary boundaries. Modern AI is a good example. It would be inaccurate to say that everything we call agentic AI today is formally neuro-symbolic AI. It is not. But the most capable systems increasingly combine things that once belonged to different intellectual worlds: neural models with search and retrieval, structured knowledge, deterministic code, tools, rules, memory, permissions, evaluators, and traditional software systems.

The progress comes from the combination. The model alone is not the system.

That brings us to the agentic Web, because one of the greatest mistakes technical people make is assuming that solving the technology also solves the business.

Very intelligent people can reason brilliantly from first principles inside their own discipline and then become strangely simplistic when they cross into business models. Technology may tell you what can exist. It does not tell you who pays for it, why they pay, how distribution works, which incentives emerge, or what happens when billions of ordinary users behave differently from the elegant model in your head.

One of my favorite examples is Google.

In 1996, Larry Page and Sergey Brin were Stanford graduate students working on a research project that became BackRub. Their key insight grew partly from academic citation analysis: a document should not be judged merely by whether it contains a keyword, but also by which other documents point toward it and how important those documents themselves are. That logic became PageRank and ultimately the foundation of Google Search.

The technology was extraordinary. But the business model was not obvious.

In their 1998 paper The Anatomy of a Large-Scale Hypertextual Web Search Engine, Brin and Page were explicitly skeptical of advertising-funded search. Their concern was legitimate: if advertisers pay the search engine, commercial incentives can conflict with the goal of returning the best possible information. They wrote that advertising-funded search engines could become “biased towards the advertisers and away from the needs of the consumers.”

Think about the irony.

The founders of what would become one of the greatest advertising businesses in history initially wrote an academic paper explaining why advertising was structurally dangerous for search.

They were not stupid.

They were not hypocrites.

And they did not suddenly become evil.

They encountered reality.

Search was expensive to build and operate. Users expected it to be free. The better Google became, the more people used it, and the more infrastructure it required. Somewhere, an economic engine had to finance that system.

At roughly the same time, Bill Gross and GoTo.com, later renamed Overture, were discovering something important: search intent had enormous commercial value. Instead of selling generic display advertising, advertisers could bid for keywords and pay when users actually clicked. GoTo pioneered the auction-driven pay-per-click model in 1998. Google initially took another route when it launched AdWords in 2000, but by 2002 it had moved to cost-per-click auctions as well. The crucial Google twist was to incorporate user response into ad ranking rather than simply allowing the highest bidder to win, helping align commercial value with relevance.

This is the part of the story I find fascinating.

Google did not resolve the conflict between advertising and search by pretending the conflict did not exist.

It redesigned the mechanism.

An irrelevant advertiser willing to pay more could not automatically deserve the best position if users consistently ignored the ad. Bid mattered, but relevance increasingly mattered too. Economics and product quality were made to reinforce each other more than they had in the earlier paid-search systems.

By the time Google went public in 2004, Larry Page could write something that would have looked almost absurd beside the 1998 paper: advertising was now Google’s principal source of revenue, while arguing that its ads could be relevant and useful rather than intrusive.

That business model financed free search and an expanding suite of consumer services at enormous scale.

Not the Internet itself. Users still paid telecom operators and ISPs for access, and publishers still financed the creation of content. But advertising allowed Google to put an extraordinarily expensive information-retrieval system in front of billions of users without charging them each time they searched.

And that economic decision helped shape the Web for the next quarter century.

This is why I become skeptical whenever a new technological paradigm begins with moral certainty about its future business model.

We hear versions of the same claim today around AI and the agentic Web: advertising was the disease of the old Web; subscriptions will create a cleaner one; perhaps micropayments will finally allow information to be priced correctly; perhaps agents will simply pay other agents per call and the whole advertising economy can disappear.

Maybe.

But consumer economics is brutally difficult.

Neeva is a useful warning. It launched with an appealing proposition: a subscription-funded, ad-free alternative to Google. Yet in 2023 its founders concluded there was no sustainable path for its consumer search business, shut the service down, and the company was acquired by Snowflake, where its technology moved into the enterprise data world.

Perplexity offers another, more nuanced example. It built around direct answers and paid subscriptions, but later began experimenting with advertising. Its own explanation was revealing: subscriptions alone did not generate enough revenue to create a scalable publisher revenue-sharing program. As of 2026, the company operates a mixture of free access, consumer subscriptions, enterprise plans, APIs, and other monetization surfaces rather than betting the company on one perfectly pure model.

None of this proves advertising wins again.

That would be exactly the kind of linear historical reasoning I am arguing against.

The agentic Web has fundamentally different economics. An agent does not need to look at a banner. It can call an endpoint directly. A machine can economically make micropayments that would be absurd for a person. Commerce can move from clicks toward transactions. Publishers can potentially meter machine access. Software can charge per call, task, or outcome. Payment protocols can make transactions almost invisible.

But someone still has to finance the system.

The models cost money.

Inference costs money.

Search and crawling cost money.

Publishers need incentives to continue creating information.

Merchants need margins.

Agent surfaces need economic engines.

And billions of consumers have repeatedly demonstrated that their theoretical enthusiasm for paying subscriptions is much greater than the number of subscriptions they actually want to maintain.

So the interesting question is not whether advertising is morally good or bad, nor whether subscriptions, micropayments, transactions, or licensing are intellectually cleaner.

The interesting question is:

What economic architecture can actually finance an agentic Web used by billions of people?

My suspicion is that the answer will be messy.

Advertising may migrate from the page into the answer or recommendation layer. Subscriptions will finance premium usage. Commerce will finance transactional agents. Enterprises will pay for productivity. Micropayments may finance individual calls and high-value information. Publishers will license some assets, expose others, fence others, and build endpoints around the rest.

In other words, the future probably will not emerge from one pristine business model.

It will emerge from the interaction of several.

And this is where history and first-principles thinking should meet.

First principles help us ask which assumptions of the old Web no longer hold.

History reminds us not to assume that the business model we find aesthetically attractive is the one reality will choose.

Google’s founders began with a technically brilliant search engine and a principled argument against the business model that ultimately financed it. Their real achievement was not merely changing their minds. It was discovering a mechanism through which the economics could be redesigned around the product.

The agentic Web is at a similar moment.

We can already see much of the technical architecture taking shape: the fact replacing the page, the call replacing part of the click, delegated identity, agent protocols, machine payments, settlement rails, provenance.

What we do not yet know is which combination of those layers will finance the system at global scale.

That is the question worth studying.

Not what the ideal Web should look like.

What business model reality will force it to become.

Access The Enterprise AI Ecosystem Atlas

If you’re already a paid member, simply reply to this email, and we’ll send it your way.

The Enterprise AI Atlas

Gennaro Cuofano — Fri, 25 Sep 2026 04:59:41 GMT

Get Premium for Full Enterprise AI Atlas

For three years, the enterprise AI market has been organized around one assumption: the model is the center of gravity.

The assumption made sense. When ChatGPT launched in November 2022, the model was the breakthrough. It packed an extraordinary amount of intelligence into a chat window anyone could use. Work that had needed specialist teams, dedicated infrastructure and narrow machine-learning systems suddenly looked like a general-purpose platform. So the first questions were about capability. Which model reasoned better, held more context, wrote better code, or ran faster, cheaper and safer? Every company added a copilot. Every software product became “AI-powered”. Every enterprise started testing model providers, private deployments, retrieval systems, and internal assistants.

Get Premium for Full Enterprise AI Atlas

Deployment exposed the limit. A model could write an excellent answer and still be nearly useless in a real operation. It did not know which customer record was authoritative. It did not understand how a contract related to an entitlement. It could not tell whether an employee could approve a refund, which exceptions needed human review, or which regulation applied to a given transaction. The model had intelligence without institutional context.

Get Premium for Full Enterprise AI Atlas

Then agents arrived, and the ambition moved from answering questions to doing work. The system would no longer recommend an action. It would open the case, update the record, contact the customer, issue the refund, escalate the exception, and document what happened.

The adoption record shows how far that ambition still is from reality. Ali Ghodsi, who runs Databricks, describes asking enterprise audiences how many of them manage fleets of agents that coordinate with each other. Almost nobody raises a hand. Most are using a chatbot as a faster search engine, plus coding assistants. His diagnosis is blunt: the frontier models are already smart enough, but they lack the context that exists inside any organisation. He sells a platform built to capture that context, so the view is not disinterested. It still matches what most deployments show. For the typical enterprise, a smarter model is not the bottleneck.

That changed the nature of the problem. Once AI begins to act, the constraint is no longer intelligence. It is context, authority, process and control. This is the market now taking shape.

The Price of Free AI

Gennaro Cuofano — Thu, 24 Sep 2026 04:59:41 GMT

One of the most important shifts over the next 12 months will be the move away from using closed frontier models for everything.

Enterprises will increasingly push more strategic workloads toward open-weight models, not simply because they can be cheaper, but because they offer something more important: modularity, control, deeper customization, and greater protection over proprietary context and intellectual property.

The strategic goal is to bring the model closer to the enterprise, rather than continuously sending the enterprise into someone else’s model. That means adapting models to internal knowledge, workflows, policies and domain-specific context, while building a deeper understanding of where the company’s real AI moat can emerge.

In theory, this is extremely attractive. In practice, running, customizing, securing and maintaining open models inside the enterprise core is far from free.

And this is where the economic argument is often misunderstood.

If the main reason you are moving to open models is simply to save money on tokens, you may be approaching the problem from the wrong direction. The stronger case for open models is control, modularity, strategic independence and the ability to progressively encode proprietary enterprise context into the AI stack.

Cost matters, but it should be evaluated against those strategic benefits, not in isolation.

This analysis breaks down that trade-off in full: where open models actually save money, where they become more expensive than expected, and where the real strategic value lies beyond the token price.

Open weights can give you more control, more flexibility and lower costs. But downloading a model, operating it and getting useful work from it are three different economic problems.

An open-weight model can cost nothing to download. That is a meaningful advantage. You receive the model’s learned parameters, with rights to use and modify them under its licence. But the download is not the service. Someone still has to supply the computing capacity, operate the software, monitor performance and handle failures. A free model does not automatically produce a low-cost system.

The opposite mistake is just as important: adding every engineering, evaluation and governance expense to self-hosting, then comparing that total with an API’s token bill alone. Buying access to a managed model does not eliminate the customer’s work on integration, access controls, output quality or oversight. It changes which responsibilities sit with the provider and which remain with the customer.

The useful question is not “Are open models cheaper?” It is “Which combination of model, deployment and supervision delivers the required outcome at the lowest acceptable cost and risk?”

That question has an established name: total cost of ownership.

The Full AI Open Ecosystem Atlas Here

To gain Full Access, subscribe below.

The New Consulting Business Model

Gennaro Cuofano — Tue, 22 Sep 2026 04:59:19 GMT

In 1933, a young lawyer named Marvin Bower joined a small Chicago consulting firm founded seven years earlier by James O. McKinsey.

Bower did not found McKinsey, nor did he invent management consulting. But he became one of the central architects of the modern profession. Harvard Business School goes further, calling him the father of modern management consulting.

Bower believed consulting should operate as a profession rather than simply another commercial trade. Over the following decades, McKinsey institutionalized many of the practices that would come to define elite consulting: the firm above the individual, an independent obligation to tell clients what consultants believed to be true, recruitment of exceptional young graduates, apprenticeship, and eventually the up-or-out system.

Together, those choices helped institutionalize one of the most successful organizational models of the twentieth century.

The pyramid.

A small number of expensive partners sat at the top. Beneath them were layers of bright, relatively inexpensive juniors who researched, interviewed, modeled, synthesized, and built the slides. The client paid for the partner’s judgment, while much of the work was produced by the pyramid underneath.

For decades, that was the economic engine of consulting: leverage.

Then AI arrived.

For seventy years, the consulting business model ran on a quiet engine: leverage.

A small number of expensive partners sat atop a large pyramid of bright, inexpensive juniors who researched, modeled, interviewed, synthesized, and built the slides. The client paid for the partner’s judgment but received, for the most part, the pyramid’s labor.

The economics followed from that ratio: billing rates, utilization, up-or-out promotions, profit per partner. A firm became more valuable when each partner could sell more junior hours beneath them.

AI attacks precisely that layer.

Research and synthesis. First-draft analysis. Public-source benchmarking. The first model. The first slide. The base of the consulting pyramid was among the first categories of professional work to become legible to a machine.

The engine did not merely become more efficient.

It was removed.

Three consequences follow:

The base became rentable. Leverage that once belonged to the firm can now be purchased by almost anyone for a subscription.
The hour begins to lose its role as the unit of value. Once clients can see the approximate cost of producing an outcome, charging for the time between input and output becomes harder.
The deliverable loses scarcity. For seventy years, the document was what the firm left behind. Now the document is cheap.

What the client increasingly wants instead is the capability.

And what the firm should have been accumulating all along is the residue of every engagement it ever ran: decisions, exceptions, cases, methods, and patterns that repeated across clients.

Instead, most of that knowledge left inside the deck.

The industry is therefore not disappearing. It is splitting between firms using AI to produce the old product faster and firms rebuilding what the product is.

This volume is for the second group.

Sell the outcome. Keep the residue. Leave the capability.

This is a new book in The Business Engineer’s Series.

The Full AI Open Ecosystem Atlas Here

If you’re already a paid member, simply reply to this email, and we’ll send it your way.

Open Escape Velocity

Gennaro Cuofano — Sun, 20 Sep 2026 11:59:33 GMT

For almost four years, the development of the AI industry followed a fairly simple assumption: the market would be dominated by closed models. OpenAI established the pattern, Anthropic reinforced it, and together they came to define much of the enterprise market. For a while, it was easy to believe that the question had been settled.

Then DeepSeek changed the picture.

The DeepSeek moment showed that a certain class of open-weight model could compete on cost, capability, and deployability in ways that mattered to enterprises. Through 2026, that opening became harder to ignore. More capable open models arrived, serving infrastructure improved, hosting became easier, and enterprises gained more credible alternatives to a purely closed-model stack.

The picture today is therefore much more nuanced. Closed models still dominate important parts of the market, particularly at the frontier. But open weights are no longer a peripheral alternative. They are becoming part of the production architecture itself.

To understand what changed, we need to separate the different layers: capability, economics, infrastructure, control, and ultimately who captures the value.

In August 2026, open-weight models processed 56% of all tokens on Vercel’s AI Gateway, up from 7% in December 2025. OpenRouter reported a similar pattern, with open models accounting for roughly 60% of US-originating token consumption during the same period. Those percentages are not market share for the entire AI industry. Gateway data sees the workloads routed through those platforms, not every direct API contract, consumer subscription, private deployment, local model, or enterprise agreement.

The reverse is also true: commercial gateways miss some open-model usage precisely because open weights can run privately. The numbers are therefore neither a census nor a curiosity. They tell us something narrower and still important: within major production environments designed around model choice, open weights have moved from experimentation into real production infrastructure.

The more important change, however, is not that open models crossed 50% of token volume. It is what has happened around those models. A model can now be trained by one organization, served by another, routed through a third, embedded inside a fourth company’s product, and specialized by a fifth for an enterprise customer. Training, serving, distribution, customization, and application delivery no longer have to sit inside the same company.

That separation is what makes the current cycle different. For years, the trajectory of open AI depended heavily on whether one or two large organizations continued releasing competitive weights. If Meta slowed Llama, or another major lab changed strategy, the open frontier could lose momentum. Today, many more actors have independent reasons to keep it moving: inference companies want workloads, chip vendors want compute demand, gateways want model diversity, application developers want substitutable intelligence, enterprises want lower costs and greater control, and regional labs want strong bases they can adapt rather than retrain from scratch.

This is what I mean by open escape velocity. Not that open models have defeated closed models. Not that every dependency has disappeared. And not that the ecosystem is already economically self-sufficient. The stronger claim is that open AI is beginning to develop enough independent sources of demand, infrastructure, capital, and institutional support that its future depends less on the strategy of any single model company. But adoption and self-renewal are different things. The gateway data tells us open models are being used. It does not yet tell us whether the ecosystem can sustainably finance the next generation of them.

The Business Engineer’s Agenting platform is now available to Executive members. You can start exploring it here: businessengineering.ai/agent

You can preview the platform at Business Engineering AI.

If you’re already a paid member, simply reply to this email, and we’ll send it your way. You can check out the full library here.

Jev & Beyond Human-in-the-Loop AI

Gennaro Cuofano — Sat, 19 Sep 2026 11:59:31 GMT

When ChatGPT launched in November 2022, what astonished me was not simply the capability of the underlying GPT‑3.5 model. It was that a powerful language model had become something people could interact with naturally. You could ask a question, challenge the answer, request a different explanation, and continue the conversation. The breakthrough was not intelligence alone. It was making that intelligence accessible through an interaction people immediately understood.

As I wrote in my newsletter at the time, an important part of that story was the work behind InstructGPT and reinforcement learning from human feedback, or RLHF. ChatGPT was a sibling of InstructGPT, not simply the same model with a chat window attached. Its training combined supervised examples of conversations with reinforcement learning guided by human rankings of responses. These methods helped turn a model trained to continue text into an assistant better able to follow instructions and respond to what a person actually wanted.

To me, that was the defining product insight: capability became approachable. The model’s intelligence mattered, but so did the removal of friction between that intelligence and an ordinary user. You no longer needed to understand the machinery underneath to get something useful from it.

Yet the same human-centered design introduces a tension. An answer a person prefers is not necessarily a decision a business should execute. Human feedback can reward correctness, clarity, and useful caution, but poorly balanced feedback can also encourage agreement, reassurance, or confidence that the evidence does not justify. OpenAI’s account of its 2025 GPT‑4o sycophancy problem illustrates the risk: a change intended to improve the experience instead made the model excessively agreeable. That is a failure mode of the incentives, not the inevitable result of all human-feedback training.

This distinction becomes more important as we move from conversational assistance to agentic execution. In a chat, a person can read the response, question it, and decide what happens next. Now imagine an enterprise operating hundreds or thousands of agent runs in parallel. Those systems may be routing requests, updating records, preparing transactions, or coordinating work across applications. A plausible explanation is no longer sufficient evidence of success. The question is whether the intended result actually occurred, not whether the agent convincingly said that it did.

If every result still requires someone to interpret and approve it, the review requirement remains part of the cost of every result. Some decisions should retain that checkpoint. But for bounded, repetitive work, the opportunity is to make review selective: let the system proceed where the evidence supports delegation, and bring the person back where uncertainty or consequences demand judgment.

That does not mean RLHF is inherently incompatible with automation. It has also improved instruction-following and truthfulness, and human feedback has been used to train agents outside conversational settings. Human involvement in training is not the same thing as requiring a human to supervise every decision in production. Simply removing RLHF would not automatically produce a more reliable system.

The more interesting question is what happens when we stop treating the conversational assistant as the default design for every application of intelligence. What would we build if the primary objective were dependable decisions, rather than preferred responses? A system designed around that objective would need to demonstrate correct task completion, uncertainty tested against outcomes, appropriate abstention, and actions constrained by explicit authority. It would not necessarily need to narrate every decision to a person.

That is where today’s issue begins: not with the assumption that removing RLHF solves automation, but with an investigation into whether automation needs a different balance of training objectives, interfaces, and evidence. The assistant had to become useful to the person reading its answer. The next system has to become dependable enough for software to act on its output.

The first breakthrough made intelligence conversational. The next may make it reliably actionable.

Jev is now available inside The Business Engineer’s Agenting platform. Executive members can start exploring it here: businessengineering.ai/agent

You can preview the platform at Business Engineering AI.

You can also experience Jev’s speed firsthand by trying Genesis, our latest free AI tool: businessengineering.ai/tools/genesis

If you’re already a paid member, simply reply to this email, and we’ll send it your way.

The Next Twelve Months of Enterprise AI

Gennaro Cuofano — Wed, 16 Sep 2026 11:59:19 GMT

One of the most striking aspects of the AI Supercycle, and one of the observations that shaped my thesis around it from the beginning, is how strange this cycle has been. It did start where most technology cycles start. It began as a consumer phenomenon. ChatGPT arrived almost as a surprise to the industry itself and immediately plugged into thirty years of existing Web distribution. Anyone with a browser could access frontier intelligence. There was no enterprise procurement cycle, no systems integration project, no six-month implementation. The distribution layer already existed.

For a while, that made AI look as if it would follow the Web’s trajectory: consumer first, enterprise second. The interface was conversational, the marginal cost of trying it was almost zero, and the first explosive use cases were individual. Even through much of 2024, the market could still be interpreted that way.

Then the shape of the competition began to change. Anthropic pushed aggressively into coding and enterprise workflows, coding agents became one of the clearest demonstrations of what sustained agentic work could look like, and the frontier labs increasingly found themselves competing not only for consumer attention but for the right to sit inside the enterprise.

OpenAI, Anthropic, Google, Microsoft, Amazon, and almost every serious model provider are now trying to solve different versions of the same problem: how do you turn general intelligence into useful work inside a very specific organization?

That shift is much more important than another model benchmark. It exposes the fundamental difference between AI and the previous wave of digitalization.

Digitalization was largely an outside-in transformation. The Web created a new external layer of distribution. Companies connected themselves to it through websites, search, e-commerce, social platforms, mobile apps, APIs, and eventually cloud infrastructure. The consumer interface changed first, and organizations reorganized themselves behind it. The Internet made the company reachable from the outside.

AI is increasingly an inside-out transformation.

It starts from intelligence and has to work its way into the machinery of the organization. And that is much harder. It does not matter how intelligent the underlying model becomes if it does not know which customer the employee means, which contract governs the transaction, which policy is current, which account is authoritative, which product can actually be sold in that market, which person owns the exception, what action it is permitted to take, and what the company considers a successful outcome.

Whether we eventually call the underlying capability AGI, ASI, or something less dramatic is almost secondary to the enterprise problem in front of us. Intelligence alone is not the product.

A frontier model has no intrinsic understanding of how your company works. It does not carry your operating model inside its weights. It does not know which version of reality your organization has agreed to treat as canonical. It does not possess authority merely because it can reason.

The bridge between intelligence and enterprise work is context.

But context here means much more than retrieval. It is the firm’s entities, relationships, rules, process states, ownership, permissions, exceptions, standards, history, and current operating reality. It is the semantic and operational layer that turns a general model into something capable of performing one narrow, vertical, consequential workflow.

The real enterprise AI problem is not simply making the model smarter. It is making intelligence situationally useful, operationally authorized, and economically valuable.

Once seen this way, the next phase of enterprise AI becomes easier to understand. For genuinely difficult work, model capability can still be the bottleneck. But for a growing class of bounded tasks, capability is already good enough that the constraint has moved elsewhere. The system needs the right context. Once it has context, it needs authority and proof. Once it can act, human attention becomes scarce. And once it operates at scale, economics decides whether it survives renewal.

This is the rotating bottleneck:

Capability → Context → Authority & Proof → Human Attention → Economics

A constraint binds until technology, capital, or organizational learning eases it. Then the next constraint becomes visible. The mistake is continuing to optimize for the bottleneck that just disappeared.

The enterprise operating loop follows naturally. Business context feeds governed execution. Governed execution produces accepted outcomes. Accepted outcomes create economic value. Exceptions generate expert judgment. Verified and permitted judgment becomes reusable learning. That learning improves the next execution and makes the next comparable deployment cheaper, faster, and more accurate. Then the loop starts again.

The weak link is that learning does not automatically belong to the vendor. Feedback can be noisy, confidential, biased, customer-specific, or contractually restricted. The loop compounds only when useful feedback can be verified, retained, and legitimately reused.

The operating doctrine for the next twelve months is therefore:

Productize deployment. Own context, control, and proof. Price accepted work.

What the Last Twelve Months Changed

In September 2025, the characteristic enterprise AI artifact was still the pilot. A team identified a use case, found an internal champion, built something impressive enough for a demonstration, and added another initiative to an innovation portfolio. In many organizations, the fact that the pilot existed was itself treated as evidence of progress.

Twelve months later, the relevant artifact is increasingly the production loop and its bill. A production loop needs a named owner. It needs access to real systems and real company data. It needs a standard against which the work can be accepted or rejected. It needs permission to act, a path for exceptions, a way to recover when something fails, and somebody willing to pay for its ongoing operation. Most importantly, it runs repeatedly. The same system that looked extraordinary in a twenty-minute demonstration now has to survive thousands of real cases.

That is where the questions change. What happens when the model is uncertain? What happens when two systems disagree? Who decides which source is canonical? Which actions can the agent execute autonomously? Which require approval? How much human review does the workflow consume? What happens when the model provider changes the underlying model? What does the whole thing cost once inference, tooling, failures, rework, and review are included?

Eventually finance asks the question that the pilot never had to answer:

What did the accepted work cost?

The rest of the market moved in parallel. Coding agents showed what an agentic workbench could look like when the environment provides strong feedback. Application vendors started opening their systems to external agents and protocols. Semantic layers, ontologies, governance systems, evaluation infrastructure, and agent control planes moved from architecture diagrams into product roadmaps. Forward-deployed engineering became a strategic capability rather than a temporary implementation function. AI engineering became more central while parts of the entry-level knowledge-work ladder began narrowing.

All of these developments are easier to understand once the cycle is framed correctly. The market did not simply move from weaker AI to stronger AI. It moved from capability to operations.

The enterprise question changed from:

Can the model do this?

to:

Can the organization operate this repeatedly?

If you’re already a paid member, simply reply to this email, and we’ll send it your way.

The Enterprise Buyer - The Book

Gennaro Cuofano — Tue, 15 Sep 2026 04:59:51 GMT

Over the last five years, I’ve been rebuilding the Business Engineer curriculum from the ground up.

The Business Engineer, a concept I first introduced in 2018 as a spin-off of my research at FourWeekMBA, had become, in my mind, the defining discipline for the digital business person. But as AI became increasingly central to the technology landscape, another shift became impossible to ignore.

A large part of that shift was about the buyer.

During the digital era, it was possible to build highly valuable software companies by obsessing over UX, narrowing the scope of the application, and distributing software to millions of small businesses and individual users. The economics made sense: standardize the product, reduce friction, and scale distribution.

AI changes that equation.

The small business may increasingly be able to vibe-code its way through much of its own software stack, assembling and customizing applications around its specific workflows rather than buying dozens of rigid SaaS products.

The enterprise sits at the opposite end of the spectrum.

Large organizations have the budgets, complexity, proprietary data, regulatory constraints, and operational depth that make customization not merely desirable, but necessary. They are willing to pay to adapt AI systems to the core of how their businesses actually operate.

Software, once again, is changing definition. I explore that transition in greater depth in the other books in this series.

But one consequence is already clear: the enterprise buyer has moved to the center of the AI adoption race.

And that is why it deserves a book of its own.

Buying enterprise AI is difficult for a simple reason: the vendor usually has far more experience selling this kind of system than you have buying it.

Imagine an insurer evaluating an AI system for claims. The claims team may buy one serious AI platform this year. The vendor may run fifty evaluations across insurers, banks and healthcare companies. It already knows which demo gets executives excited, which objections procurement will raise, which security questions arrive late, which pilot metrics are easy to pass and how far it can move on price before the buyer walks away.

The buyer has different expertise. The claims leader knows claims. Security knows security. Procurement knows contracts. Finance knows the budget. But nobody in the room may have seen twenty claims-AI deployments go from impressive demo to messy production.

That matters because AI is unusually easy to make look good before production. Suppose the vendor demonstrates the system on twenty claims it selected itself. Eighteen are handled correctly. The demo looks excellent. But those twenty cases tell you almost nothing about what happens when a scan is missing a page, two policy clauses conflict, a claimant gives contradictory information, the retrieval source is stale, the underlying model changes next month, or the case falls into the strange five percent that consumes half of your adjusters’ time.

Traditional software was easier to inspect. If you bought a CRM, you could verify that the workflow existed, permissions worked, the integration connected and users could complete the required actions. AI behaves differently. A system can work on the vendor’s examples and fail on yours. It can change after a model upgrade. It can become expensive when power users begin using it seriously. It can also improve from the cases, corrections and decisions your own people provide.

So change the order of the purchase. Do not begin with the demo. Begin with your work.

If you are buying an AI system for claims, give every shortlisted vendor the same real claims and have your senior adjuster define what correct looks like. If you are buying an AI support agent, test it on the tickets your best support people consider difficult. If you are buying an AI contract system, include broken scans, unusual clauses, amendments, missing pages and the exceptions that actually consume lawyers’ time.

Then measure the result yourself.

The vendor should not choose the exam, write the answers and grade the paper.

Your company needs to own the exam.

The same principle applies after purchase. Imagine that after six months the claims team has produced eighty original test cases, two hundred failures found in production, hundreds of adjuster corrections, escalation rules and a record of which decisions led to successful recoveries. That material now describes how your company performs part of the work.

If it all lives inside the vendor’s system, leaving becomes expensive. If you own it in portable form, the vendor becomes much easier to replace.

Three questions should therefore sit underneath every material AI purchase:

Does it work on our cases?
What do we become dependent on?
What do we still own after a year of using it?

Everything else follows from those questions. The demo earns the right to be tested. The pilot produces evidence. The contract decides ownership and change rights. The renewal asks whether the evidence still justifies the dependency. The exit plan matters because a clause saying you may leave is useless if leaving requires six engineers and nine months of rebuilding.

Own the test. Own what the system learns from your work. Know what it would cost to leave.

If you’re already a paid member, simply reply to this email, and we’ll send it your way.

Context & Graph Engineering - The Book

Gennaro Cuofano — Mon, 14 Sep 2026 04:59:40 GMT

For more than a decade, I have been fascinated by neural networks for a simple reason: they felt intuitively powerful because they were, however loosely, inspired by the architecture of the human brain. The caveat matters. Artificial neural networks are not faithful simulations of biological brains. They began as biologically inspired computational abstractions and have since evolved largely according to the demands of mathematics, data and computation.

Part of that fascination came from neuroscience itself. I spent part of my early professional life in San Diego, home to an unusually dense neuroscience ecosystem around institutions such as UC San Diego and the Salk Institute. In my early twenties, one of the ideas that struck me most deeply was neuroplasticity: the realization that the brain is not a fixed structure, but something continuously reshaped by experience, learning and the environment. Experiences do not simply fill the brain with information. They can alter its functional organization and even its physical structure.

That idea stayed with me.

Fast-forward to the mid-2010s. I was sitting in a startup office trying to understand where AI might go next. By then, neural networks were no longer merely a promise. Deep learning had already produced major breakthroughs in computer vision, speech recognition and sequence modeling. Yet these systems still seemed remarkably limited when compared with the messiness of intelligence in the real world.

This was one reason I found reinforcement learning so compelling.

AlphaGo became perhaps the most visible demonstration. Its famous Move 37 against Lee Sedol in 2016 looked almost alien to expert Go players, yet emerged from a system combining deep neural networks, search and reinforcement learning. Around the same period, reinforcement learning also occupied a prominent place in OpenAI’s early research agenda, with projects such as OpenAI Gym and Universe focused explicitly on agents learning through interaction with environments.

What appealed to me about reinforcement learning was its relationship with experience. Intelligence in the real world is not simply a matter of performing reasoning in isolation. We act, encounter consequences, update our understanding, and act again. The world gives us feedback. Much of what we eventually call intuition is accumulated experience compressed into a form that allows us to navigate situations whose relationships are nonlinear, incomplete and difficult to express explicitly (by the way this is what The Business Engineer’s discipline is all about).

And this is where the story becomes particularly interesting.

The extraordinary progress of modern AI has come primarily from neural methods. Yet neural models have an enduring problem: they do not encounter a task with all of the context that makes that task meaningful in the real world. They can contain enormous amounts of learned statistical knowledge, while still lacking the specific entities, relationships, rules, history, constraints and state required to operate reliably inside a particular environment.

This is where symbolic structures become important.

Neuro-symbolic AI refers broadly to attempts to combine the learning and pattern-recognition strengths of neural networks with explicit representations of knowledge and reasoning. Knowledge graphs, ontologies, rules and other structured representations can become part of that architecture. They do not replace the neural model. They give it something it otherwise struggles to construct reliably on demand: an explicit map of the world in which it is expected to act.

That is the subject of this book.

Context and Graph Engineering is about engineering that missing layer: how we represent entities, relationships, semantics, memory, rules and situational context so that increasingly powerful neural models can become increasingly useful in the real world.

The neural network gives us extraordinary generalization.

The graph gives us structure.

Context tells the system what world it is operating in.

And increasingly, useful intelligence will come from making those pieces work together.

This is another volume in The Business Engineer’s Foundation Series.

If you’re already a paid member, simply reply to this email, and we’ll send it your way.

The New Roles of the AI Economy

Gennaro Cuofano — Sat, 12 Sep 2026 11:59:30 GMT

In 1935, the word computer did not primarily describe a machine.

It described a person.

At Langley Memorial Aeronautical Laboratory, the predecessor of NASA’s Langley Research Center, a young mathematician named Virginia Tucker joined four other women in the laboratory’s first computing pool. Their job was to calculate. They processed experimental data, solved equations, plotted results and produced the mathematical work engineers needed to design aircraft.

They were called computers.

Over the following decades, hundreds of women would do this work at Langley. During World War II, the demand became so large that NACA expanded its computing workforce further, including the Black women of the segregated West Area Computing unit whose story would later become widely known through Hidden Figures.

Then the machines arrived.

And something interesting happened.

The work did not simply disappear. It moved.

The machine increasingly performed the calculation. The human moved toward programming it, checking it, interpreting its output and deciding what problem it should solve. Some of the women who had been called computers became among the early programmers of the electronic-computing age. Dorothy Vaughan, who had led Langley’s West Computing group, became an expert FORTRAN programmer as NASA reorganized around electronic computing.

The title migrated from the person to the machine.

The responsibility migrated in the opposite direction.

That history is useful because something similar is happening again.

The strongest signal in the AI labor market is not the proliferation of strange new titles. It is a change in who owns what.

Commercial teams are hiring builders. Product managers are taking responsibility for evaluation. Lawyers and other domain experts are moving from reviewing software to helping construct and supervise it. Deployment is splitting into engineering, rollout and enablement. AI work is moving directly into finance, recruiting, marketing, customer service and operations.

Meanwhile, much of the work that historically trained the entry-level professional — research, first drafts, basic analysis, annotation, ticket handling — is becoming the machine’s work.

Once again, the interesting question is not simply which jobs disappear?

It is:

Where does the human move when the machine takes the task?

This volume is an attempt to map that movement.

Across eleven business functions, forty-four kinds of work are already becoming visible. Their titles remain inconsistent, but the underlying pattern is remarkably stable. Humans are moving away from simply producing the work and toward designing the systems that produce it, supplying the expertise those systems need, evaluating what they produce, handling the exceptions they cannot resolve, and owning the outcomes they create.

The computer once moved from the person to the machine.

Now something similar is happening to knowledge work.

And once again, the work is not simply disappearing.

It is moving.

The shape of the new job can be compressed into three responsibilities:

Own a loop. Sit at a junction. Seed the standard.

This is another deep dive in The Business Engineer’s Foundation Series.

If you’re already a paid member, simply reply to this email, and we’ll send it your way.

The AI Business Valuation Book

Gennaro Cuofano — Fri, 11 Sep 2026 04:59:11 GMT

One of the subjects I loved most during my business studies was corporate finance. I was fortunate to have excellent professors, and one reason I eventually landed at the University of San Diego School of Business in my early 20s was the opportunity to study with Thomas Copeland, one of the scholars whose work helped shape modern thinking on corporate finance and valuation.

I became so fascinated with the discipline that I made it a major emphasis of my MBA. At the time, I had a relatively simple belief: if you could truly master corporate finance, you could build an extraordinary professional advantage. Finance appeared to offer something close to a science of business, a rigorous framework through which almost any company, investment, or strategic decision could ultimately be understood.

Little did I know that corporate finance would teach me almost the opposite lesson.

Finance is extraordinarily powerful precisely because it deals with uncertainty. But that also means it can never be the hard science I once imagined it to be. Its models are abstractions of a world that is nonlinear, reflexive, path-dependent, and filled with variables that cannot be known in advance.

That tension became one of the intellectual foundations behind The Business Engineer.

One reason I started writing about business was to help more people understand a lesson I had first encountered through finance: a model is useful because it helps you think, not because it gives you the truth.

A discounted cash flow is a perfect example. It can be an extraordinarily powerful instrument. But there is no serious DCF without a view on the discount rate, and no serious estimate of the cost of equity without assumptions about things such as the risk-free rate and the equity risk premium. Of course, you can simply use whatever rate has become conventional in the market. But then what exactly are you analysing?

The arithmetic may be precise while the underlying assumptions remain deeply uncertain.

That is not a weakness of corporate finance. It is actually where its greatest usefulness begins.

Corporate finance gave me a toolbox that forced me to connect the micro and the macro: the economics of an individual company with interest rates, capital markets, risk, incentives, competitive dynamics, and expectations about the future. Properly used, those tools expand your thinking. They force you to expose assumptions, test scenarios, and make explicit what would otherwise remain intuition.

The problem begins when the toolbox becomes the worldview.

Too often, financial models are treated not as instruments for thinking but as representations of reality itself. Once that happens, the apparent precision of the model can hide the uncertainty of its assumptions. And some of the largest mistakes in business come from exactly this inversion, both on the upside and the downside: companies that look worthless because a model cannot capture what is compounding beneath the surface, and companies that look extraordinarily valuable because a spreadsheet extrapolates temporary economics indefinitely into the future.

I say this as someone who genuinely idolised finance.

As a teenager, one of my intellectual heroes was Federal Reserve Chairman Alan Greenspan. Years later, as the 2008 financial crisis propagated through the global economy and into the 2010s, I followed Ben Bernanke’s explanations of the crisis and monetary system with enormous attention, including his 2012 lectures on the Federal Reserve and the financial crisis.

I was fascinated by the idea that sufficiently sophisticated financial and economic models might allow us to understand an extraordinarily complex system.

Over time, however, finance itself taught me the limits of that belief.

The deeper lesson of the financial crisis, and of many of the business cycles that followed, was not that models were useless. It was that a discipline becomes dangerous when it develops the presumption that its models can explain the whole world.

Finance is not the worldview.

It is one tool inside a much larger toolbox.

That toolbox also needs technology, strategy, economics, history, psychology, organisational behaviour, geopolitics, systems thinking, and the accumulated judgment that comes from watching real businesses behave in ways no spreadsheet predicted.

The goal is not to abandon models. It is to combine them into a stronger first-principles approach: use models to decompose reality, expose assumptions, identify mechanisms, and understand what would need to be true for a thesis to hold, without ever confusing the model with reality itself.

Finance was my first intellectual love in business, and for that reason I will probably never stop loving it.

But I can no longer reconcile myself with the limitations it creates when used carelessly, particularly when tools designed for one economic regime are carried unquestioningly into another.

This book in the Business Engineer Foundation Series is dedicated to that problem.

To taking the extraordinary toolbox corporate finance gave us, understanding where it still works, identifying where its assumptions have broken, and rebuilding it for a world in which the assets, economics, capital structures, and companies themselves are changing faster than the models traditionally used to value them.

The purpose of finance should not be to shrink reality until it fits the model. It should be to give us better tools for thinking about a reality that will always remain larger than the model.

If you’re already a paid member, simply reply to this email, and we’ll send it your way.

The AI Deployment Architect - The Book

Gennaro Cuofano — Mon, 07 Sep 2026 04:59:08 GMT

Back in 2023, when almost everyone was caught off guard by the first AI wave, one thing was already clear to me: the enterprise had moved to the center of the AI story. The question was no longer whether large companies would adopt AI. It was whether they would accept what adoption really implied: AI would eventually force a change to the core of the organization itself.

Fast-forward to 2026, and most enterprises have, in one form or another, crossed the adoption threshold. This year they have spent aggressively on models, tokens, copilots, agents, and the infrastructure required to run them, racing from experimentation toward agentic operating models. In that sense, adoption itself is no longer the main constraint.

And that is precisely what makes the current moment so revealing. Once usage becomes abundant, the real bottleneck becomes visible. It is not access to intelligence. It is not even willingness to use it. The hard part is translating all of that intelligence into deployed business outcomes across an organization. A company can have thousands of employees using AI every day, consume enormous amounts of inference, launch dozens of pilots, and still struggle to show where margin improved, where a workflow fundamentally changed, or where a new competitive capability was created.

I have spent much of my career around enterprise software, working at the intersection of customers, technical teams, and commercial organizations, and this is what feels fundamentally different about AI. With previous software waves, adoption was often the hard part: get the system installed, get people into it, get usage up. With AI, usage can spread almost by itself. The friction appears one level higher, when the organization tries to turn thousands of local productivity gains into one coherent operating model.

Who decides which workflows should be automated? Where does human judgment remain? Which models and agents can act on which data? Who owns the corrections, evaluation sets, and decision history they generate? How does one successful experiment become a company-wide standard rather than another isolated tool?

Those are governance questions. And governance has traditionally been treated as something that comes after innovation: first let teams experiment, then bring in security, legal, procurement, architecture, and the operating model once something starts working. AI turns that sequence upside down. Governance is not the layer that follows the transformation. It is the mechanism that allows the transformation to compound.

Without it, the enterprise can generate enormous activity at the edges while accumulating very little at the center. Every employee gets faster. Every team produces more. Every dashboard turns green. Yet the company can still end up in roughly the same competitive position, because its rivals bought access to the same intelligence at roughly the same time.

That is the tension this piece is trying to unpack. The enterprise AI challenge is no longer principally one of adoption. It is one of organization: how individual gains become coordinated systems, how local discoveries become standards, how the knowledge generated by humans and agents stays with the firm, and how all of it is ultimately aimed at a strategic position competitors cannot reproduce simply by buying the same models.

The transformation therefore does not end when the enterprise adopts AI.

That is where the difficult part begins, and this book comes in handy.

This is a new book in The Business Engineer’s Foundation Series.

If you’re already a paid member, simply reply to this email, and we’ll send it your way.

The Agentic Organization - The Book

Gennaro Cuofano — Sat, 05 Sep 2026 11:01:48 GMT

I have been working with AI in one form or another since around 2015, when much of what we called AI in practice was still narrow machine learning and natural-language processing. Back then, getting machines to understand language meant stitching together specialized libraries, classifiers, entity extraction, rules, and carefully structured data.

Around 2017 and 2018, I remember working on ways to make websites “talk”: turning their content into structured data that machine interfaces and voice assistants could actually consume. It worked, but it was incredibly clunky. You could see the direction of travel, but much of what we imagined still felt decades away.

Then the underlying technology began to change. The arrival of transformer-based models, Google bringing systems such as BERT into search, and the rapid evolution of the GPT family between 2018 and 2020 made it increasingly obvious that we were crossing into something different.

These systems were becoming dramatically more general. Instead of building a separate narrow pipeline for every language task, you could increasingly place different problems in front of the same underlying model.

At first, I was not particularly impressed by the outputs themselves. As an author, analyst, and writer, I could still do things these systems simply could not do well. The important signal for me was elsewhere: the rate at which the boundary was moving.

Things that a few years earlier I had mentally placed ten or twenty years into the future were suddenly becoming plausible within a product cycle. The clunky interfaces we had spent years engineering around narrow NLP were beginning to collapse into much more general systems.

By the early 2020s, a large part of my day-to-day work had consequently become figuring out how to bring this new generation of AI into the enterprise stack. And that is where I started to see a second problem emerge.

The hard part was increasingly not getting the technology to do something impressive. It was getting the impressive thing through the long journey from prototype to production, and then from one production deployment to something that could actually scale across an organization.

Back in 2023, when almost everyone was caught off guard by the first major generative-AI wave, one thing was therefore already clear to me: the enterprise had moved to the center of the AI story. The question was no longer simply whether large companies would adopt AI. It was whether they would accept what adoption really implied. AI was not going to remain another piece of software layered onto the existing organization. Eventually, it would require a change to the core of the organization itself.

Fast-forward to 2026 and most enterprises have, in one form or another, crossed the adoption threshold. This year they have spent aggressively on models, tokens, copilots, agents, and the infrastructure required to run them, racing from experimentation into the agentic era. Enterprises have consumed enormous amounts of intelligence. Employees are using AI. Developers are building with it. Business units are launching agents. In that sense, usage is no longer the main constraint.

And that is precisely what makes the current moment so revealing. Once usage becomes abundant, you can finally see where the real bottleneck sits. It is not access to intelligence. It is not even willingness to use it. It is the translation of all that intelligence into deployed business outcomes across the enterprise.

A company can have thousands of employees using AI every day, spend heavily on inference, launch dozens of pilots, and still struggle to answer the questions that ultimately matter. Where did margin improve? Which workflow was fundamentally redesigned? What can the organization now do that it could not do before? Which capability is compounding rather than simply making an employee faster?

I have spent much of my professional life at the intersection of customers, technical teams, and commercial organizations, and this is what feels fundamentally different from previous software waves. Historically, adoption itself was often the hard part: buy the system, integrate it, get people into it, drive usage. With AI, usage can spread almost by itself. The friction appears one level higher, when the company attempts to turn thousands of local productivity gains into one coherent operating model.

Who decides which workflows should actually be automated? Where does human judgment remain? Which agents may take which actions? Which models may touch which data? Who owns the evaluation sets, corrections, and decision history those systems generate? What happens when a model changes underneath a production workflow? How does one successful experiment become a company-wide standard rather than the forty-third isolated AI tool inside the company?

These are fundamentally governance questions. Yet governance is almost always treated as the thing that comes after innovation. First experiment. First prove something. First let teams move quickly. Then bring in security, legal, procurement, architecture, data governance, and the operating model once the technology has demonstrated its value.

I think AI turns that logic upside down.

Governance is not the layer that follows the transformation. It is the mechanism through which the transformation compounds.

Without it, the enterprise can produce extraordinary activity at the edges while accumulating remarkably little at the center. Every employee becomes faster. Every department launches agents. Every productivity dashboard turns green. Yet the knowledge generated by those systems remains fragmented, the workflows remain structurally unchanged, and the company can end up in approximately the same competitive position because its rivals bought access to the same intelligence in the same quarter.

That is the tension this piece is trying to unpack. The enterprise AI challenge is no longer principally one of adoption. It is one of organization: how individual gains become coordinated systems, how local discoveries become institutional standards, how the knowledge generated by humans and agents remains with the firm, and how all of this is ultimately aimed at a strategic position competitors cannot reproduce simply by purchasing the same models.

The technology moved faster than I expected when I was working with those narrow NLP systems a decade ago. But that has only made the organizational problem more visible.

The transformation does not end when the enterprise adopts AI. That is where the difficult part begins.