Has AI actually grown businesses, or just adoption

Two years in, the honest question is not whether you use AI. It is whether any of it changed the number at the bottom.

Technology
By Mark Choudhari · Jun 2, 2026 · 6 min read

Everyone adopted AI. The business stayed flat. That gap is Phase 4.
Made with Works

TL;DR

Around year two with AI, the question changes from whether to adopt it to whether any of it moved the business. Adoption is near-universal and material value is rare. The Phase 4 question is that honest accounting, and the data answers it the same way for most founders: busier with AI, not better because of it.


In this article

Somewhere around year two with AI, the question changes. It stops being “should we adopt this” and becomes “did any of it actually move the business.” That second question is the one that ambushes you, and for most founder-led businesses the answer is uncomfortable. The hype promised a transformed P&L. The data shows adoption almost everywhere and material value almost nowhere. The felt version is quieter and worse: you are clearly doing more with AI, and the business runs about the same.

This is not a story about founders getting it wrong. Adopting AI was the reasonable move, the one the whole market made. The gap is between what the tools promised and what they delivered, and naming that gap is the whole point of this piece, because the gap is the only thing left to close.

Has AI actually grown businesses, or is AI delivering ROI

Mostly not, and the firms that count these things agree on the shape of it. Adoption is near-universal while value is rare. The promise of 3x growth and self-running functions ran straight into a P&L that did not move, and the measured reality is now well documented. Everyone said adoption was the hard part. Adoption already happened. The results did not follow it.

The adoption side of the ledger keeps filling in: Gallup found frequent AI use in the workplace continuing to rise quarter after quarter through the end of 2025, to the point where AI now runs somewhere inside nearly every organization. The value side never kept pace. Only a minority of companies can point to any profit-and-loss impact from AI, most have not begun scaling it across the enterprise at all, and the small group creating substantial value at scale keeps pulling further away from the wide majority generating none. Put adoption and value next to each other and the conventional wisdom inverts.

Why isn't AI improving my revenue

Because using AI and getting a business result from it are two different things, and most of the effort is still stuck at the first one. A tool that drafts faster, summarizes faster, or automates a single step does not, on its own, change revenue or margin. Worse, a lot of AI activity is quietly adding work rather than removing it.

This is the productivity paradox, and it is measured. HBR documented “workslop,” the plausible-looking AI output that someone downstream has to catch and redo, with an estimated invisible cost of around $186 per month per worker. A controlled METR study found that experienced developers were roughly 19% slower with AI tooling than without it, while believing they were faster.

The plausible-looking AI output someone downstream has to catch and redo carries an estimated cost of around $186 per month per worker.
Harvard Business Review, 2025

The felt experience of AI, faster and more productive, and the measured experience, flat or slower, have come apart. It helps to see why. Think of AI in a business as three layers: the tools at the bottom, the tasks they speed up in the middle, and the business results at the top. Almost all the AI most businesses bought operates at the bottom two layers. It makes tasks faster. It does not move the top layer, because making a task faster is not the same as the business doing better. Efficiency went up. Revenue did not. (We go deeper on this in the Three-Layer Pyramid.)

What is the real productivity impact of AI in 2026

Real, but mostly at the task level, and mostly invisible at the business level. Frequent workplace AI use roughly doubled across the workforce, from about 12% to about 26% of workers using it at least a few times a week, so adoption is genuinely up. What did not move in step is output that shows up in the P&L. Everyone adopted. Few benefited.

The honest version is that AI made individual tasks faster while leaving the operating system of the business untouched. A founder can produce a plan in twenty minutes that used to take two days and still end the quarter with the same revenue, because the plan was never the bottleneck. The handoffs were. The stalled work was. The fact that nothing shipped end to end without the founder pushing it was. Task-level AI does not touch any of that, which is why the productivity gains are real in the moment and invisible in the accounts. This is the gap behind every one of the named-firm numbers above. The tools improved before the operating layer did, so the business felt the speed without ever booking the result.

What the Phase 4 question actually asks

It is one question, asked honestly. After two years of experiments, did the business get better because of AI, or did it just get busier with AI? The data says, for most, busier. Near-universal adoption, a minority with any bottom-line impact, a majority with no material value at all. The accounting is not flattering, and pretending otherwise helps no one.

But this is a question, not a verdict, and the reason the experiments did not move the business is fundamental, not personal. Experiments are not operations. More pilots do not close the gap, because the gap was never a shortage of pilots. What closes it is moving AI into the actual operating rhythm of the business, against the real work, with results you can point to. That is the shift from experiments to operations, and it is the answer the question is asking for.

The counter-argument deserves a fair hearing: it is early, and value compounds. Fair. The timeline concedes. What does not concede is the mechanism. The businesses closing the gap are not the ones running more pilots. They are the ones scaling AI into operations.

The shift Phase 4 is asking for, and how Works fits

If the question is whether AI moved the business, then the answer it demands is not another experiment. The bar Phase 4 sets is specific: the AI has to run work end to end against the real business and report a result you can take to a board. That is an operations bar, not an experiment bar, and it is the bar Works is built to clear.

JynAI built Works, an AI Business OS, to be the layer that turns experiments into operations. Here is the fit, plainly.

  • Pain: every task is faster and the number is flat.
    Work That Actually Ships runs the whole process in three modes, Strategy to plan, Action to execute across the tools you already use, Automation to run hands-free, so the work finishes rather than staying stuck between one tool and the next.
    Gain: the result comes out the top instead of motion staying stuck at the bottom.

  • Pain: there is no way to tell what the AI actually did.
    Receipts logs every run, every agent action, and every outcome, and rolls them up at the area and workspace level, exportable to a board deck.
    Gain: the accounting Phase 4 is asking for now has an answer, and it is not a weekend spreadsheet.

  • Pain: the same people evaluating new AI tools are also the people who need the business to run.
    Expert-Grade Workflows arrives with 500+ plays built on EOS, MEDDIC, ABM, and PLG, so there is no blank page and no shortlist to maintain. The plays come pre-built; the deciding has already happened inside the system.
    Gain: the evaluation cycle that was eating your week stops being your job.

  • Pain: the AI investment does not compound because every new model or tool means starting over.
    Keeps Getting Better holds 100+ models in the pool and auto-selects per step. When a new frontier model ships it joins the pool and your existing work uses it without you touching a thing.
    Gain: the reset that used to land on you every few months stops being your problem, and the investment accumulates rather than resets.

The affordability is the part that makes this honest. The full capability set is available at the $49 tier, not behind an enterprise contract, so the founder who would have spent another quarter in Phase 4 can reach operations without a budget line that requires a committee to approve.

Machintel spent close to two years at Phase 4, running experiments that produced activity and nothing worth taking to a board. Once the operations layer was in place, six teams were running on it in ninety days. We are biased about our own product. The argument underneath is not: if the experiments did not move the business, more experiments were never going to, and the gap they left was always an operations gap. Phase 4 is not a permanent state. It is the question. Phase 5 is the answer.

Get the answer Phase 4 demands. Sign up for early access. Or see the data behind the value gap in the AI Value-Gap Data Cut.

Common Questions

Has AI actually grown businesses?

For most, not materially. Adoption is near-universal while bottom-line impact is rare. McKinsey found that 88% of companies now use AI, but only about 6% are actually generating substantial value at scale. The businesses closing that gap are not the ones running more pilots. They are the ones that moved AI into operations.

Why isn’t AI improving my revenue?

Because using AI and getting a business result from it are two different things, and most AI activity is still stuck at the task layer. A tool that makes one task faster does not, on its own, change revenue or margin. The gap is between the task layer and the outcome layer, and that gap does not close by adding more tools at the task layer. What closes it is the operating layer that runs the whole process, with the handoffs handled, and is accountable for whether the result shipped.

What is the real productivity impact of AI in 2026?

AI’s productivity impact in 2026 is real at the task layer and largely absent at the business layer. Adoption doubled from roughly 12% to 26% of workers using AI several times a week, yet the P&L did not follow. Two measured counterweights explain the gap: HBR put the “workslop” rework cost at $186 per worker per month, and METR clocked experienced developers running 19% slower with AI tools than without while reporting they felt faster. More adoption, less result.

What is the Phase 4 question?

Phase 4, Clarity, is the point where a founder stops and asks one honest question: did the business actually get better because of AI, or did it only get busier? It is the fourth of five phases in the AI buyer’s journey, and the first one that measures results rather than activity. Most who reach it cannot answer it. McKinsey data puts 88% adoption and roughly 6% at real scale, which means Phase 4 discomfort is the majority experience, not a personal failing.

How do you move from AI experiments to AI operations?

You move when the AI runs the work end to end against the real business without a person standing behind it, and when a result comes out the top that you can point to. The mechanism is an operations layer above the tools and tasks, the layer that was skipped when every business went straight from chat tools to a stack of point solutions. That layer is not a better tool. It is a different kind of thing entirely: the layer that runs the work rather than doing a step of it.

Recent Posts
AI Economic Growth
The DIY Tax and what building AI yourself really costs
Getting AI working is one cost. Keeping it working is the tax.
The Discovery Tax and what choosing AI really costs
Integration Tax
The Integration Tax and why your AI keeps breaking

Has AI actually grown businesses, or just adoption

Two years in, the honest question is not whether you use AI. It is whether any of it changed the number at the bottom.

Technology
By Mark Choudhari · Jun 2, 2026 · 6 min read

Everyone adopted AI. The business stayed flat. That gap is Phase 4.
Made with Works

TL;DR

Around year two with AI, the question changes from whether to adopt it to whether any of it moved the business. Adoption is near-universal and material value is rare. The Phase 4 question is that honest accounting, and the data answers it the same way for most founders: busier with AI, not better because of it.


In this article

Somewhere around year two with AI, the question changes. It stops being “should we adopt this” and becomes “did any of it actually move the business.” That second question is the one that ambushes you, and for most founder-led businesses the answer is uncomfortable. The hype promised a transformed P&L. The data shows adoption almost everywhere and material value almost nowhere. The felt version is quieter and worse: you are clearly doing more with AI, and the business runs about the same.

This is not a story about founders getting it wrong. Adopting AI was the reasonable move, the one the whole market made. The gap is between what the tools promised and what they delivered, and naming that gap is the whole point of this piece, because the gap is the only thing left to close.

Has AI actually grown businesses, or is AI delivering ROI

Mostly not, and the firms that count these things agree on the shape of it. Adoption is near-universal while value is rare. The promise of 3x growth and self-running functions ran straight into a P&L that did not move, and the measured reality is now well documented. Everyone said adoption was the hard part. Adoption already happened. The results did not follow it.

The adoption side of the ledger keeps filling in: Gallup found frequent AI use in the workplace continuing to rise quarter after quarter through the end of 2025, to the point where AI now runs somewhere inside nearly every organization. The value side never kept pace. Only a minority of companies can point to any profit-and-loss impact from AI, most have not begun scaling it across the enterprise at all, and the small group creating substantial value at scale keeps pulling further away from the wide majority generating none. Put adoption and value next to each other and the conventional wisdom inverts.

Why isn't AI improving my revenue

Because using AI and getting a business result from it are two different things, and most of the effort is still stuck at the first one. A tool that drafts faster, summarizes faster, or automates a single step does not, on its own, change revenue or margin. Worse, a lot of AI activity is quietly adding work rather than removing it.

This is the productivity paradox, and it is measured. HBR documented “workslop,” the plausible-looking AI output that someone downstream has to catch and redo, with an estimated invisible cost of around $186 per month per worker. A controlled METR study found that experienced developers were roughly 19% slower with AI tooling than without it, while believing they were faster.

The plausible-looking AI output someone downstream has to catch and redo carries an estimated cost of around $186 per month per worker.
Harvard Business Review, 2025

The felt experience of AI, faster and more productive, and the measured experience, flat or slower, have come apart. It helps to see why. Think of AI in a business as three layers: the tools at the bottom, the tasks they speed up in the middle, and the business results at the top. Almost all the AI most businesses bought operates at the bottom two layers. It makes tasks faster. It does not move the top layer, because making a task faster is not the same as the business doing better. Efficiency went up. Revenue did not. (We go deeper on this in the Three-Layer Pyramid.)

What is the real productivity impact of AI in 2026

Real, but mostly at the task level, and mostly invisible at the business level. Frequent workplace AI use roughly doubled across the workforce, from about 12% to about 26% of workers using it at least a few times a week, so adoption is genuinely up. What did not move in step is output that shows up in the P&L. Everyone adopted. Few benefited.

The honest version is that AI made individual tasks faster while leaving the operating system of the business untouched. A founder can produce a plan in twenty minutes that used to take two days and still end the quarter with the same revenue, because the plan was never the bottleneck. The handoffs were. The stalled work was. The fact that nothing shipped end to end without the founder pushing it was. Task-level AI does not touch any of that, which is why the productivity gains are real in the moment and invisible in the accounts. This is the gap behind every one of the named-firm numbers above. The tools improved before the operating layer did, so the business felt the speed without ever booking the result.

What the Phase 4 question actually asks

It is one question, asked honestly. After two years of experiments, did the business get better because of AI, or did it just get busier with AI? The data says, for most, busier. Near-universal adoption, a minority with any bottom-line impact, a majority with no material value at all. The accounting is not flattering, and pretending otherwise helps no one.

But this is a question, not a verdict, and the reason the experiments did not move the business is fundamental, not personal. Experiments are not operations. More pilots do not close the gap, because the gap was never a shortage of pilots. What closes it is moving AI into the actual operating rhythm of the business, against the real work, with results you can point to. That is the shift from experiments to operations, and it is the answer the question is asking for.

The counter-argument deserves a fair hearing: it is early, and value compounds. Fair. The timeline concedes. What does not concede is the mechanism. The businesses closing the gap are not the ones running more pilots. They are the ones scaling AI into operations.

The shift Phase 4 is asking for, and how Works fits

If the question is whether AI moved the business, then the answer it demands is not another experiment. The bar Phase 4 sets is specific: the AI has to run work end to end against the real business and report a result you can take to a board. That is an operations bar, not an experiment bar, and it is the bar Works is built to clear.

JynAI built Works, an AI Business OS, to be the layer that turns experiments into operations. Here is the fit, plainly.

  • Pain: every task is faster and the number is flat.
    Work That Actually Ships runs the whole process in three modes, Strategy to plan, Action to execute across the tools you already use, Automation to run hands-free, so the work finishes rather than staying stuck between one tool and the next.
    Gain: the result comes out the top instead of motion staying stuck at the bottom.

  • Pain: there is no way to tell what the AI actually did.
    Receipts logs every run, every agent action, and every outcome, and rolls them up at the area and workspace level, exportable to a board deck.
    Gain: the accounting Phase 4 is asking for now has an answer, and it is not a weekend spreadsheet.

  • Pain: the same people evaluating new AI tools are also the people who need the business to run.
    Expert-Grade Workflows arrives with 500+ plays built on EOS, MEDDIC, ABM, and PLG, so there is no blank page and no shortlist to maintain. The plays come pre-built; the deciding has already happened inside the system.
    Gain: the evaluation cycle that was eating your week stops being your job.

  • Pain: the AI investment does not compound because every new model or tool means starting over.
    Keeps Getting Better holds 100+ models in the pool and auto-selects per step. When a new frontier model ships it joins the pool and your existing work uses it without you touching a thing.
    Gain: the reset that used to land on you every few months stops being your problem, and the investment accumulates rather than resets.

The affordability is the part that makes this honest. The full capability set is available at the $49 tier, not behind an enterprise contract, so the founder who would have spent another quarter in Phase 4 can reach operations without a budget line that requires a committee to approve.

Machintel spent close to two years at Phase 4, running experiments that produced activity and nothing worth taking to a board. Once the operations layer was in place, six teams were running on it in ninety days. We are biased about our own product. The argument underneath is not: if the experiments did not move the business, more experiments were never going to, and the gap they left was always an operations gap. Phase 4 is not a permanent state. It is the question. Phase 5 is the answer.

Get the answer Phase 4 demands. Sign up for early access. Or see the data behind the value gap in the AI Value-Gap Data Cut.

Common Questions

Has AI actually grown businesses?

For most, not materially. Adoption is near-universal while bottom-line impact is rare. McKinsey found that 88% of companies now use AI, but only about 6% are actually generating substantial value at scale. The businesses closing that gap are not the ones running more pilots. They are the ones that moved AI into operations.

Why isn’t AI improving my revenue?

Because using AI and getting a business result from it are two different things, and most AI activity is still stuck at the task layer. A tool that makes one task faster does not, on its own, change revenue or margin. The gap is between the task layer and the outcome layer, and that gap does not close by adding more tools at the task layer. What closes it is the operating layer that runs the whole process, with the handoffs handled, and is accountable for whether the result shipped.

What is the real productivity impact of AI in 2026?

AI’s productivity impact in 2026 is real at the task layer and largely absent at the business layer. Adoption doubled from roughly 12% to 26% of workers using AI several times a week, yet the P&L did not follow. Two measured counterweights explain the gap: HBR put the “workslop” rework cost at $186 per worker per month, and METR clocked experienced developers running 19% slower with AI tools than without while reporting they felt faster. More adoption, less result.

What is the Phase 4 question?

Phase 4, Clarity, is the point where a founder stops and asks one honest question: did the business actually get better because of AI, or did it only get busier? It is the fourth of five phases in the AI buyer’s journey, and the first one that measures results rather than activity. Most who reach it cannot answer it. McKinsey data puts 88% adoption and roughly 6% at real scale, which means Phase 4 discomfort is the majority experience, not a personal failing.

How do you move from AI experiments to AI operations?

You move when the AI runs the work end to end against the real business without a person standing behind it, and when a result comes out the top that you can point to. The mechanism is an operations layer above the tools and tasks, the layer that was skipped when every business went straight from chat tools to a stack of point solutions. That layer is not a better tool. It is a different kind of thing entirely: the layer that runs the work rather than doing a step of it.