The full stack world

Most organisations turn an idea into a working thing by passing it down a chain. A strategist frames the problem, a designer gives it form and an engineer builds it. Between each of them sits a layer of people whose job is to move the work from one specialism to the next: the brief, the account lead, the business analyst, the project manager, the steering committee and the status report.

The chain exists for good reasons. Each specialism took years to learn, and few people could practise more than one of them to a professional standard. Ronald Coase described the logic in 1937. A firm brings work inside its walls when coordinating it there costs less than buying it on the open market.⁵ The same reasoning applies inside the firm. Every handover between specialisms has a cost, and organisations created roles to absorb it.

Most of us at Artist spent years inside large consulting programmes, and we have watched how the chain behaves at scale. A strategy is written as a deck. The deck becomes a brief for a design team, the designs become requirements for a delivery team, and the requirements become tickets. Each step is a translation, and each translation loses some of what the person before it meant. By the time the work reaches the person building it, the reasoning behind it has often been reduced to a list of acceptance criteria.

Large language models, and the tools built on them, have lowered the cost of working in a specialism that is not your own. A strategist can produce a working prototype. A designer can ship production code. An engineer can draft the positioning and test it with customers before writing the first line of a feature.

The evidence is early, and most of it comes from controlled tasks rather than whole organisations. It points in one direction. In a randomised experiment, developers using GitHub Copilot finished a set coding task 55.8 percent faster than developers without it.¹ A study of professional writers published in Science found that access to ChatGPT cut the time spent on writing tasks by 40 percent and raised the quality of the output as judged by evaluators.² When Harvard Business School and Boston Consulting Group gave consultants access to GPT-4, they completed 12.2 percent more tasks, 25.1 percent faster and at higher quality, on work inside the model’s capabilities.³

Software made the same move once before

Software development went through an earlier version of this compression. Under the waterfall method, a project moved through requirements, design, build, test and release in sequence, and each phase was signed off before the next began. Winston Royce, whose 1970 paper is usually cited as the origin of the method, wrote in that paper that the sequence was “risky and invites failure”, because testing came last and was the first point at which the design met real conditions.⁶ In 2001, seventeen practitioners published the Agile Manifesto, which valued “responding to change over following a plan”.⁷ Agile teams release small working versions to real users every few weeks and change course on what they learn.

The same reasoning applies to strategy. A strategy is a theory about how customers, competitors and the organisation will respond to a set of choices. Until it is implemented in the world and used by real people, nobody can predict its full effect or the consequences nobody intended. Mike Tyson made the point before a fight in 1987, in a line now usually remembered as “everyone has a plan until they get punched in the face”.⁸

Agile shortened the distance between a plan and its first contact with users inside the build team. The layers above the build team stayed where they were. Strategy was still written as a deck and handed down, so an organisation learned how the market received its strategy only after every handover was complete. AI extends the compression upwards. When the person who writes a strategy can build a working version of it within days, the strategy meets real users while it is still cheap to change.

Translation gets cheaper than coordination

The effect on structure follows from Coase. When one person can move from an idea to a design to a working version in an afternoon, the handovers disappear, and so does much of the case for the roles that managed them. The layer that turned a strategy into a brief, and the brief into tickets, was paid to carry intent between people who could not otherwise reach each other. A model now carries a large share of that load, inside a single person’s working day, where intent does not have to be written down to survive.

Startups show the result first because they carry no legacy structure. In March 2025, Y Combinator reported that a quarter of its Winter 2025 batch had codebases in which 95 percent of the code was generated by AI.⁴ We see the same shape in our own work. On an Artist engagement, the people who frame a client’s problem are the people who build the system that solves it, and there is no separate strategy phase handed over to a delivery phase.

The cost of compression

Compression has costs, and they deserve the same attention as the gains. A specialist with ten years in one discipline still sees things that a generalist with a model will miss. In the Harvard and Boston Consulting Group study, consultants who used GPT-4 on a task outside its capabilities were 19 percentage points less likely to reach the correct answer than consultants who worked without it.³ A person who can now produce work in four disciplines can also produce confident, plausible work in four disciplines without the judgement to see where it is wrong.

Review becomes more important as relay disappears. The organisations that benefit will keep deep specialists close to the work, as reviewers and as the people who set the standard, and remove the layers that only moved the work between them.

When the layers between strategy, design and execution compress, the work concentrates on the people at either end of the old chain: the people who decide what is worth making and the people who make it.

Visionaries and entrepreneurs

A visionary sets a direction other people want to follow. As the cost of building falls, a clear view of what should exist becomes the scarce input, and a visionary can now test that view by building a working version instead of describing it and waiting for someone else to interpret it.

An entrepreneur takes on risk to turn an opportunity into a business. The capital and headcount needed to reach a first product have fallen, which lowers the cost of each bet and raises the number of bets one person can make. The constraint moves from funding a team to choosing the right problem.

Value creators and problem solvers

A value creator is measured by the outcome they produce: revenue won, cost removed, a customer problem solved. In a compressed organisation the line from a person to a result is shorter and easier to see, and roles that cannot show that line become harder to justify.

A problem solver stays with a problem until it is solved, across whatever disciplines it crosses. These are the people the chain frustrated most, because it asked them to stop at the edge of their specialism and hand the problem on. A model lets them keep going.

Builders

A builder makes the working thing. Building now takes in the strategy and the design, because a builder with a model can hold all three. This is how Artist hires and how we work, and it is the reason the company exists.

The layers in between

The roles under the most pressure are those whose value lies in moving work between other people: turning a deck into a brief, a brief into requirements, requirements into tickets and progress into status reports. Many of the people in those roles have the context and judgement to move to either end of the chain, and the same tools that compress the layers make that move easier. An account manager who understands the client can build the prototype. A project manager who knows where delivery goes wrong can become the reviewer who catches it.

For an organisation, the practical question is where its own layers sit. Map the path from a decision to a working thing and count the handovers. For each one, ask whether it exists to carry information that one person with a model could now carry themselves. The answer will differ by function, and the deep specialists should stay. The handovers are where the time and the money go.