The Workflow Dissolution Effect

The champagne was flat by the time he noticed. Not literally. But the celebration should have landed differently.

His operations manager had just shown him the numbers: their new AI workflow system was processing client onboarding in fourteen minutes instead of three hours. The team had spent six months building it. It worked beautifully.

Then she asked the question he had been avoiding. “So what does the onboarding coordinator do now?”

He looked at Sarah’s desk. Sarah, who had been with them since they were a five-person operation. Sarah, whose job title still existed. Sarah, whose actual work had just evaporated.

Not her tasks. Her tasks were still there, scattered across other people’s workflows. The workflow itself, the coherent sequence of decisions that justified a full-time human making them, had dissolved.

This is the part most founders are not seeing.

The Wrong Conversation

The debate about Artificial Intelligence (AI) and jobs has split into two camps, and both are wrong.

Camp one says AI replaces jobs. They point to the numbers: 1.2 million positions eliminated globally in the past twelve months. Block cutting over 4,000 roles, roughly 40% of their workforce, with Chief Executive Officer (CEO) Jack Dorsey predicting that “the majority of companies will reach the same conclusion within a year.” The raw statistics are real. The framing is not.

Camp two says AI creates more jobs than it destroys. They cite LinkedIn’s data showing 1.3 million new AI-related roles since 2023. They quote Jensen Huang telling workers they are “confusing your job with the tools you use to do it.” This is also real. And also beside the point.

Both camps are arguing about the wrong unit of analysis.

AI does not replace jobs. AI does not augment workers. AI dissolves workflows.

Here is what that means, and why the math is worse than either camp realises.

What Workflow Dissolution Actually Looks Like

Think about what a “job” actually is in a company with five to twenty-five people. It is not a single task repeated. It is a bundle of workflows, and each workflow is a sequence of decisions.

Your operations coordinator does not “do operations.” She processes invoices (a workflow containing about fifteen decisions per invoice: verify amounts, match purchase orders, flag discrepancies, approve payment). She schedules meetings (a workflow containing about eight decisions: check calendars, assess priority, determine duration, send invitations). She manages vendor relationships (a workflow containing dozens of judgment calls about reliability, pricing, and quality).

Task automation would make her faster at each of those things. Maybe she processes invoices in two minutes instead of ten. She is still your operations coordinator. She still justifies a full-time salary.

Workflow dissolution is different. When agentic AI absorbs the entire invoicing workflow, the scheduling workflow, and the routine vendor communications, what remains is a scattered collection of judgment calls: the vendor relationship decisions that require context, the invoice discrepancies that require investigation, the scheduling conflicts that require human diplomacy.

Those judgment calls might take two hours per day. They used to be embedded inside eight hours of workflow execution. Now they stand alone. And two hours of judgment work does not justify a full-time salary.

The job title survives. The headcount does not.

The 40% Restructuring Threshold

A study published on arXiv in April 2026 extending the Acemoglu-Restrepo task exposure framework puts numbers on this pattern. The researchers calculated that for each percentage point increase in job automation, there is a 0.75 percentage point actual job displacement. But the relationship is not linear.

Below the 40% mark, augmentation works. AI handles some decisions within the workflow, the human handles the rest, and the structure holds. The economics favour keeping the same people doing enhanced work.

Above 40%, the economics flip. It becomes structurally simpler to redesign the workflow with AI as the primary executor and humans as exception handlers, rather than maintaining the current architecture with AI bolted onto human workflows.

43% of current occupations have already crossed this threshold. Not in five years. Now.

Below 40%, you are augmenting. Above 40%, you are restructuring whether you planned to or not.

The median projection from the research: 9.3 million American jobs are vulnerable to AI-driven displacement within two to five years. The range spans from 2.7 million under slower adoption to 19.5 million if agentic AI adoption accelerates. Perhaps the most concerning figure: 4.9 million workers across 33 occupations sit at the tipping point, projected to swing from under 10% to over 40% displacement within that window.

These are not factory workers. These are operations coordinators, compliance processors, customer support teams, and mid-level managers. The people who run the workflows that run the business.

The Containerisation Parallel

This has happened before.

Before 1956, shipping worked like this: crews of up to twenty-five longshoremen would manually load and unload cargo piece by piece. The process took days. Each worker’s job was a bundle of workflows: sorting, carrying, stacking, securing. Task-level automation came first. Forklifts and pallets made individual loading faster. Each longshoreman became more productive.

Then Malcolm McLean introduced the shipping container. He did not automate the task of loading a crate. He dissolved the workflow of piece-by-piece cargo handling entirely. The container replaced the entire sequence of decisions about how to sort, stack, and secure individual items. What remained was a fundamentally different job: operating cranes that moved standardised containers.

The displacement was not incremental. Port districts that contained major ports in 1971 showed significantly lower employment twenty years later, according to London School of Economics (LSE) research. Forty years later, the gap persists.

The structural parallel is exact. Task automation in knowledge work is the forklift. Agentic AI workflow automation is the container. The question is not “which tasks will AI do?” but “which of your workflows will survive as intact workflows?”

The Founder’s Unique Vulnerability

If you run a company with five to twenty-five people, this affects you differently than it affects a Fortune 500 executive.

In large organisations, the architect-executor split already exists. Managers design workflows. Workers execute them. When AI dissolves the execution layer, the architects remain. The restructuring is painful but structurally familiar: redeploy, retrain, reduce through attrition.

In your business, you are both architect and executor. You designed the client onboarding workflow AND you execute parts of it yourself. Your operations manager designed the invoicing process AND processes invoices. Your marketing lead designed the content calendar AND writes the content.

AI is forcing a split you have never had to make. For the first time, you need to decide: am I the person who designs how this business operates, or the person who operates it?

Enterprise has been preparing for the architect-executor split for decades. The founder-operator is confronting it for the first time.

The Block restructuring illustrates the enterprise version. Dorsey’s internal AI platform, reportedly codenamed “Goose,” combines large language models with workflow automation. It fetches data, drafts code, routes support tickets. Managers reported 30% improvement in response times with stable quality. The result: 4,000 positions eliminated, $450-500 million in restructuring charges, and analyst margin forecasts raised by 250 basis points.

Block can absorb that restructuring. They have 6,000 remaining employees, redeployment programmes, and a restructuring budget larger than most small businesses’ annual revenue.

You have twelve people. Each one is load-bearing. When a workflow dissolves, you cannot redeploy anyone to a department that does not exist. You face a binary: redesign your entire decision architecture or continue operating a structure that your competitors are about to make obsolete.

Your Business Is a Decision Architecture

Here is the reframe that changes the math.

Stop thinking about your business as a collection of jobs. Start thinking about it as a collection of decisions.

Some decisions are routine. They follow the same logic every time. Process this invoice. Schedule this meeting. Send this reminder. Route this ticket. These decisions are embedded inside workflows, and the workflows are assigned to people. But the decisions themselves do not require a human. They require a rule applied consistently.

Some decisions require judgment. Should we take this client? Is this vendor’s quality declining? Does this strategic direction serve our long-term positioning? Is this employee struggling or developing? These decisions require context, experience, values, and the kind of pattern recognition that comes from years of navigating similar situations.

Your workforce should be sized to the judgment decisions. Not to the job titles.

This is not about “cutting staff.” It is about recognising that the current organisational structure, people assigned to job-title-sized bundles of workflows, is an artefact of an era when human execution was the only way to process routine decisions. That era is ending.

The Decision Architecture Audit

Here is the practical framework. You can run this on your business in a single afternoon.

Step 1: Map your workflows. Not your job descriptions. Your workflows. “Client onboarding” is a workflow. “Monthly financial close” is a workflow. “Content production” is a workflow. List them all. Most businesses with ten to twenty employees have between fifteen and forty distinct workflows.

Step 2: For each workflow, list the decisions it contains. An invoice processing workflow contains decisions about verification, matching, approval, exception handling. A hiring workflow contains decisions about sourcing, screening, interviewing, offer construction. Be specific. “Process invoices” is not a decision. “Determine whether this invoice amount matches the purchase order within acceptable variance” is a decision.

Step 3: Classify each decision. Routine: same logic applies in most cases, can be expressed as a rule, requires data but not judgment. Judgment: requires context, experience, values, or the kind of pattern recognition that cannot be reduced to a rule.

Step 4: Calculate your exposure. For each workflow, what percentage of its decisions are routine? Any workflow above 40% is a restructuring candidate. Not because you must restructure it today, but because the economics now favour redesigning it around AI execution with human exception-handling.

Step 5: Identify the judgment clusters. The judgment decisions do not distribute evenly across your current roles. Some people make many judgment calls. Others execute workflows that are predominantly routine. The judgment clusters show you what your future team looks like: the people who make the decisions that matter.

Step 6: Design your architect roles. The people who remain are not “doing the same job with AI help.” They are workflow architects: designing how the business operates, making judgment calls across multiple former roles, and managing AI systems that execute the routine sequences. This is a fundamentally different job. It requires different skills, different thinking, and, often, different people.

The Offensive Case

Everything above might sound defensive. Audit your exposure. Identify your vulnerabilities. Prepare for the restructuring that is coming.

Here is the part that changes the calculation entirely.

The founder-operator who redesigns their decision architecture first does not just save money. They build a structural advantage that compounds. Research from the Harvard Business Review (HBR) from February 2026 found that AI does not reduce work. It intensifies it. The surviving workers do more, at higher quality, across a broader scope. One marketer doing the strategic work of three. One analyst managing what a small team used to handle.

For a GP-scale business, this means the first mover does not just reduce costs. They operate at a fundamentally different level of capability. Their twelve-person team, restructured around decision architecture, produces the output of a thirty-person team organised around job titles.

Your competitors cannot match this by “adopting AI.” They can only match it by redesigning their own decision architecture. And that requires the same uncomfortable split you have already made.

The asymmetry is this: the restructuring is painful to do, but it is painful exactly once. The competitive advantage compounds for years. The founders who wait will eventually face the same restructuring, but from a position of competitive disadvantage.

What This Is Not

This is not a call to fire your team. It is a call to understand the architecture of your business at the level of decisions rather than job titles.

Some of what you discover will be uncomfortable. You may find that a role you thought was critical is actually a workflow execution position where 60% of the decisions are routine. That does not mean the person is dispensable. It means the role needs redesigning, and the person needs reskilling or redeployment into the judgment work that actually requires them.

Some of what you discover will be reassuring. You may find that your team makes far more judgment calls than you realised, that your business is more human-dependent than the macro statistics suggest, that your exposure is lower than the industry averages.

Either way, you will know. And knowing is the difference between designing the next version of your business and having it designed for you by competitive pressure.