When Individual Speed Outruns Shared Understanding

Individual speed reaches the customer when verification, dependencies and shared understanding keep pace.

An organisation delivers when work from different people becomes a result someone can rely on.

A faster draft helps. So does a faster code change. The gain reaches the customer when the surrounding work can absorb it: assumptions are checked, dependencies are resolved and the result meets the Commitment that justified doing the work.

In The Physics of Organizations, I described how the quality of a Commitment depends on the shared understanding behind it. AI makes the cost of maintaining that understanding easier to see. When production accelerates, the time spent waiting for clarification or correcting an assumption occupies more of the journey.

The question is where the time went and what would have to change for the whole journey to improve.


What the Productivity Figures Measure

Software engineering offers a useful place to examine this question because changes leave a record that can be studied.

Yegor Denisov-Blanch, now a research scientist at the Stanford Artificial Intelligence Laboratory, studies AI use and engineering output. In a January 2026 interview, he described research involving more than 100,000 developers. His team evaluates code changes against expert assessments and uses models to extend those assessments across repositories.

He reported a median improvement of around 10 to 15 percent in their output measure, with substantial variation between teams. He also emphasised a limit: the measurements describe changes in output; explaining those changes requires further investigation. They do not measure the throughput of an entire company.

McKinsey’s 2026 survey approaches the question through respondents’ reports. Eighty percent said AI had improved their individual productivity. Thirty-seven percent attributed some positive enterprise EBIT impact to AI. Those percentages describe different measures. They cannot tell us how much of an individual speed gain survived into delivery or establish what caused the difference.

Together, these sources sharpen a question I have been working on: under what conditions does faster individual work improve an organisation’s ability to keep its Commitments?


Where the Work Goes

Imagine a team preparing a new pricing offer. Sales needs the offer for a customer conversation. Product describes the intended behaviour. Engineering implements it. Finance checks how the resulting charges will be recorded.

AI helps each person prepare their part quickly. The specification is complete by lunch. The code and tests are ready the following morning.

One assumption remains unresolved. The specification treats a change of billing schedule as a change in the service purchased. The tests inherit that assumption because they were generated from the same document. Finance discovers the discrepancy when reviewing the first invoices.

The time saved in drafting now has to be considered alongside the time spent correcting the offer, the implementation and the customer records. Work that looked complete returns to the people who produced it.

This example illustrates a mechanism. It does not establish that every AI deployment follows it. Where requirements are clear and verification is dependable, faster production can translate into faster delivery. Where assumptions cross organisational boundaries without being checked, additional output can increase the queue of work waiting for judgement.

The same distinction matters when measuring progress. A team needs to know how much work was completed, how long it waited between steps and how often it had to return. A count of generated drafts cannot answer those questions.


Changing the Work Between People

Adding another approval step may catch the billing error. It also adds a recurring cost. The more useful investigation starts with why the relevant knowledge was absent when the proposal was formed.

Perhaps the rule was already documented and difficult to find. Perhaps its owner was unclear. Perhaps the customer exception had never reached the people implementing the offer. Each explanation suggests a different change.

Peter van Sabben makes a related argument in Introducing the Rebuilder: organisations can optimise existing tasks while leaving the assumptions of their operating model untouched. His distinction helped me separate the work that creates customer value from the work required to carry information between existing departments and tools.

Some handoffs protect an essential boundary. An independent review can expose an assumption that the author cannot see. Other handoffs exist because someone has to copy a decision from one system into another.

Removing that copying can release capacity. Removing the independent judgement can make a Commitment more fragile. The organisation needs to understand which kind of work it is changing.


How Decisions Connect

This brings the question back to authority.

Strategy gives direction. The Board governs matters within its mandate. Executive leadership allocates resources and establishes operating boundaries. Teams make decisions close to the work within the authority they have been given.

The connection runs in both directions. A team that discovers an impossible delivery assumption must be able to bring that evidence back to the people responsible for the promise. A revised priority must reach the people whose work depends on it.

This is the sequence I want our systems to support. Each decision should make clear what others may rely on and what remains open. People should be able to act within their mandate without reconstructing the whole organisation around every question.

In The Rebuild Has No Finish Line, van Sabben proposes measuring decision latency and distinguishing reversible decisions from decisions with lasting consequences. That distinction matters. Reducing every decision to a speed target would reward skipping the thought some decisions require.

For the pricing example, useful measures would include the time from a proposed offer to a verified customer response, the work returned for correction and the frequency of exceptions discovered after release. They connect speed to what the organisation promised to deliver.


The Shared Understanding Behind Delivery

A shared model is the organisation’s current account of the situation: the relevant definitions, evidence, assumptions and Commitments. People can hold different perspectives within it. They need enough common understanding to act coherently and enough visibility to recognise disagreement.

Maintaining that model has a cost. Definitions need owners. Exceptions need to be recorded. Changes need to reach the people who depend on them. That cost belongs in any assessment of whether a new way of working improves delivery.

AI can help with this work as well as with production. It can bring relevant evidence into a discussion, compare proposals and identify inconsistencies for review. Whether it helps must be judged across the journey from question to delivered outcome.

The next essay follows the same problem into a company where software agents act alongside people. When many systems can act at once, differences in their understanding become differences in what the organisation does.


Core thesis: Individual speed improves delivery when the organisation can verify the work, resolve dependencies and carry a consistent Commitment through to its outcome.

The idea you can't unsee: Time saved in drafting can reappear as waiting and correction elsewhere in the same journey.

Vocabulary shift: "individual productivity" → "delivery against a Commitment"

Connects to: Article 1 (When AI Progress Becomes Hard to Question), Article 3 (When a Hundred Agents Hold a Hundred Versions of the Business), Article 1 of The Organizational Capability (The Cost of Coordination)

Version: 1.1 / 2026-11-07