Contents
A web developer changes where their effort goes
The same starting point
A client needs a website that serves its users and business goals. An experienced developer brings knowledge of discovery, design, implementation and testing.
Previously, the project moved through discovery, requirements, wireframes, design options, functionality design, development, testing and launch. Each stage involved client presentation, feedback and approval, sometimes over several rounds. Much of the developer's own time went into coding.
What changes when AI becomes part of this work?
Three ways to arrange the work
Same starting situation and goal in each. Switch at any point; the labels describe arrangements, not a ranking.
Redesign for amplification: invest expertise earlier
Based on the practitioner accountThe developer reorganizes the work around detailed preparation and rapid implementation. Human effort concentrates on the decisions that guide AI action across the project.
- The person
- Establishes goals, users and consequential constraints
- AI
- Helps structure questions and research needs
- Result or handoff
- A shared problem definition
- The person
- Judges relevant examples and resolves trade-offs
- AI
- Audits examples and drafts a detailed requirements document and design rationale
- Result or handoff
- A specification explaining both what to build and why
- The person
- Discusses a separate AI's critique and decides which changes to keep
- AI
- A separate AI reviews the plan; the original AI helps revise it
- Result or handoff
- A stronger plan, still subject to human judgment
- The person
- Checks scope, assumptions and phase gates
- AI
- Produces development instructions, a phased build plan and required tests
- Result or handoff
- Clear instructions and checks before each next stage
- The person
- Tests consequential behaviour and evaluates usefulness
- AI
- Implements the plan, runs checks and revises the build
- Result or handoff
- Working software for human and client review
- The person
- Interprets feedback and decides the necessary changes
- AI
- Helps revise the plan and implementation quickly
- Result or handoff
- Another working version for review as needed
Why more steps can take less time
The expensive parts of the previous workflow (research, requirements work, coding and testing) can move much faster. The person spends more effort making the direction sound. Working software and revisions can then arrive sooner, changing when some client review is useful.
Early agreement on important goals still matters. Fast execution also makes it possible to follow a mistaken assumption quickly.
Open a decision: what expertise contributes
A client asks for “a simple booking form.” The developer recognizes that duplicate bookings would be a serious failure. They specify how availability must be checked when a booking is submitted, what happens if two people request the same slot, and what the user sees when a slot is no longer available.
AI uses those requirements to plan the implementation and tests. The developer tests the conflict case. Domain knowledge has shaped many lines of code and checks before those lines are produced.
A wider choice of implementation
The practitioner reports previously choosing among a small number of familiar approaches. AI makes less familiar approaches more practical to consider. The developer can give more weight to project fit while still judging maintainability, accessibility and support.
The reported result and the remaining constraint
Reported outcome · not independently evaluatedThe developer reports that their own portion of work, which once took months, can sometimes now take days.
That does not mean the entire project finishes in days. Other constraints often prevent overall completion from becoming much faster. The local change is substantial; its effect on the wider project is a separate question.
| View | What the account supports |
|---|---|
| Developer's contribution | A reported large reduction in the time needed to complete their own work |
| Overall project | Other constraints may still determine when the project can finish |
| Additional organizational benefit | Needs investigation: how is the released capacity used, and what limits it? |
Why this matters to the policy
This account gives us a configuration worth studying. Which parts depend on the developer's expertise? How large is the quality-adjusted gain? What can another developer reproduce? What stops the organization from using the extra capacity? [§4.1] [§4.5] [§8.3]
Those are the questions a measurement and learning system is intended to investigate. [§6]