Knowledge work is changing in one specific direction: less doing, more framing and verifying. The organizations that understand this early will compound value. The ones that don't will mistake motion for progress.

TL;DR
  • The constraint comes first. Point AI at a broken workflow and you just go broke faster.
  • Work is shifting from doing to framing and verifying. The people who adapt become conductors of value.
  • The 10-80-10 method: spend 10% framing the problem, let AI do 80% of the heavy lifting, then spend 10% validating the output.
  • AI amplifies talent. It does not replace it. Fix your workflow first, then integrate.

The Constraint Problem Nobody Wants to Talk About

Most businesses approach AI the same way they approached every other productivity tool. They buy it, roll it out, and wait for results. And for most of them, results don't come.

The IMF estimates that 60% of jobs in advanced economies will materially change because of AI. The World Economic Forum puts 92 million jobs at risk of displacement while projecting 170 million new roles created. These are enormous numbers. But they obscure a more immediate question for business owners: why are the majority of organizations getting worse results after adopting AI tools?

The answer is the constraint. When you automate or accelerate a broken process, you do not fix the process. You scale the dysfunction. A contracting firm that quotes inaccurately and then adds an AI estimating tool does not quote more accurately. It quotes inaccurately, faster and at higher volume.

Studies on AI-assisted knowledge work suggest productivity lifts of around 40% are achievable. The operative word is achievable. Organizational research tracking AI adoption shows a split: roughly 94-96% of organizations that grab AI tools see flat or negative returns in the near term. The remaining 4-6% compound value. The difference is not the tool. It is what the tool is pointed at.

Call the first group the J-curve organizations. They dip before they rise, and many never rise. Call the second group the K-curve organizations. They move differently from the start, because they fix the constraint before they integrate the tool.

What "Finding the Constraint" Actually Means

A business constraint is the single bottleneck limiting your throughput. It is rarely where you think it is.

A cladding and contracting firm came to us with a growth ceiling they couldn't explain. Revenue was solid but stuck. Their instinct was that they needed more salespeople. After mapping the actual workflow, the constraint was estimating. Quotes were taking too long, contained errors, and required senior time that should have been going elsewhere. The firm doubled revenue in six months, not by hiring more salespeople, but by rebuilding the estimating process and then integrating AI into the redesigned workflow.

That sequence matters. Redesign first. Integrate second.

The firms that skip step one and go straight to step two are the J-curve organizations. They are moving faster toward the wrong destination.

Finding your constraint means asking a specific set of questions:

  • Where does work pile up or slow down?
  • Where does quality degrade under volume?
  • Where is senior time being consumed by tasks that shouldn't require it?
  • Where do you lose deals, clients, or revenue, and why?

The answers point at your constraint. That is where AI integration belongs.

The Shift From Doing to Framing and Verifying

Here is the structural change happening underneath all of this. Knowledge work used to reward people who could produce outputs. Write the report. Build the model. Draft the proposal. Generate the estimate. Volume and speed were human advantages.

AI changes the value equation. A capable AI system can produce a first-draft report, a financial model, a proposal, or an estimate faster than any human. What it cannot do reliably is understand the business context well enough to frame the right question in the first place. And it cannot be held accountable for the output.

That accountability, and that contextual judgement, stays with humans. And it becomes more valuable, not less, as AI handles more of the production work.

The people who thrive in this shift are not the ones who work the hardest. They are the ones who get clearest on what outcome is actually needed, give AI a tight enough brief to get useful output, and have the judgement to catch what's wrong with the output before it goes out the door.

This is framing and verifying work. It is different from doing work. It requires a different set of skills, a different mindset, and in many organizations, a different org chart.

The 10-80-10 Method

A practical framework for this shift is what we call 10-80-10.

10% framing. Before anything else, you define the problem clearly. What output do you need? What constraints apply? What does good look like? What would make this output unusable? This is human work. It requires business context, client knowledge, and stakes awareness that AI doesn't carry.

80% AI heavy lifting. With a clear frame, you hand the production work to AI. Drafting, researching, calculating, structuring, synthesising. This is where the productivity lift lives. A task that took a day can take an hour. A task that took an hour can take five minutes.

10% verification. A human reviews the output against the original frame. Not just a quick read. A deliberate check: Is this accurate? Does it answer the right question? Are there errors that would matter? Does it reflect the context AI didn't have access to? This is where judgement earns its keep.

The 10-80-10 method does not mean humans do less important work. It means humans do more important work, and they do more of it in less time.

From Individual Contributors to Conductors of Value

The role change here is meaningful. In traditional knowledge work, the most valuable person in a team was often the one who could produce the most. The fastest coder. The most prolific writer. The analyst who could build a model overnight.

In a hybrid human-agent environment, the most valuable person is the one who can direct, orchestrate, and quality-control the work of multiple AI agents while keeping everything tied to real business outcomes. That is a conductor role. And it is not the same skill set as being a strong individual contributor.

Think about what a conductor actually does. They do not play the instruments. They understand each instrument's capabilities, they know the score, and they make real-time judgements about how the whole thing should come together. They are accountable for the result.

That is what high-value knowledge work looks like in a team that uses AI well.

A therapy clinic we worked with grew from 3 practitioners to 16 over a defined period. The growth was not driven by the practitioners seeing more clients. It was driven by removing the administrative and coordination work that was consuming clinical time, and then building systems that let a small team manage a much larger practice. The practitioners became conductors of their caseload rather than manual processors of their admin.

A construction firm moved from $42 million to $180 million in revenue. Not by working harder. By rebuilding the constraint points in their workflow, integrating AI into the redesigned processes, and retraining their people to work at a higher level of abstraction. Their project managers stopped being paper processors. They became decision-makers.

Do It With Your People, Not To Them

The single biggest implementation mistake we see is organizations that treat AI integration as a technology project rather than a people project.

The technology is not the hard part. The hard part is getting your team to work differently. And people do not change how they work because a tool was installed. They change how they work when they understand why the change matters, when they are involved in designing the new process, and when they can see how the change makes their day better rather than threatening their role.

When you redesign a workflow and integrate AI without involving the people who do the work, you get resistance, workarounds, and poor adoption. The tool sits unused or underused. The investment does not pay off.

When you bring your team into the process, something different happens. They know where the pain points are. They have ideas about what good output looks like. They become advocates for the new way of working rather than resistors. And they build the contextual knowledge to do framing and verification well.

AI amplifies talent. That is not a platitude. It is a precise statement about how the technology works. If you have people who understand your business, your clients, and your standards, AI makes them dramatically more productive. If you have people who don't, AI makes their gaps more visible at higher speed.

The investment in people is not separate from the investment in AI. It is the same investment.

What Leadership Needs to Do Differently

If your organization is moving toward a framing-and-verifying model of work, leadership has to model it first.

That means being explicit about what good framing looks like. It means setting standards for verification, not just output. It means rewarding people for catching errors and improving processes, not just for producing volume. And it means being direct about which roles will change significantly and helping people navigate that change.

The role of a manager in a hybrid human-agent team is not the same as in a traditional team. Managers need to understand what AI can and cannot do, so they can set realistic expectations in both directions. They need to be able to evaluate the quality of a brief, not just the quality of an output. And they need to build a culture where framing and verifying is taken as seriously as producing.

This is a capability gap in most organizations right now. It is also one of the highest-leverage places to invest.

The K-Curve Path

The organizations compounding value from AI are not more sophisticated than yours. They are not better funded. They are not earlier adopters of every shiny tool.

They do three things differently.

First, they find the constraint before they integrate anything. They map the workflow, identify where throughput is actually limited, and redesign the process before they add AI to it.

Second, they train their people to work at a higher level of abstraction. Framing and verifying work is a learnable skill. It requires practice, feedback loops, and explicit standards. The K-curve organizations build those systems deliberately.

Third, they treat implementation as an ongoing process rather than a one-time project. The first integration reveals new constraints. The team learns what good framing looks like and gets better at it. The verification step catches new categories of errors and improves the AI's brief as a result. The system gets smarter because the people running it are getting smarter.

This is the compounding dynamic. It is not automatic. It is the result of deliberate choices made consistently over time.

The J-curve organizations buy the tool and wait. The K-curve organizations redesign the work and then use the tool to do it faster and better.

Frequently asked questions

Framing means defining the problem, the output you need, and the criteria for success before AI touches the task. Verifying means a human reviews the AI output against that original frame with real judgement, not just a quick read. Together they are the bookends of productive AI-assisted work.
Look for where work piles up, where quality degrades under volume, and where senior people are doing tasks that shouldn't require their level of experience. That bottleneck is your constraint. Fix the process at that point before you add any AI tools to it.
AI amplifies the talent you already have. People who develop strong framing and verification skills become more valuable, not less, as AI handles more production work. The risk is not replacement. The risk is irrelevance for people who don't adapt to working at a higher level of abstraction.
Recap

Work is shifting from doing to framing and verifying, and the organizations that adapt first will compound advantage over the ones that don't. The 10-80-10 method gives your team a concrete structure for the shift: frame the problem tightly, let AI carry the production load, then verify the output with genuine judgement. The constraint in your workflow has to come first. Point AI at a broken process and you scale the dysfunction. Fix the process, integrate the tool, and bring your people along as designers of the new way of working.

Your next action: map one workflow in your business this week and identify where throughput is actually limited. That is your starting point.