Skip to main content
Why DoCoDeGo · now

The Bottleneck Moved

Agile served us well — and many DoCoDeGo principles inherit directly from it. But when AI handles implementation in minutes, the bottleneck shifts. Here is what changes, why, and the eight specific frictions teams are already feeling.

The bottleneck has moved

Agile organised work around human coordination. Sprint cycles, standups, retrospectives — designed to keep humans aligned when implementation was the bottleneck.

DoCoDeGo organises work around intent clarity. When AI handles implementation, the coordination bottleneck disappears. What remains is ensuring that what is being built is what was intended.

This page is not an attack on Agile. It is an honest catalogue of the eight specific places Agile strains under AI, with the framework's evolution for each.

Eight concrete challenges

The eight specific frictions.

Each card names a real friction, cites a 2025–26 source, and points to the specific pillar, statute, or metric in DoCoDeGo that answers it.

01 Pillar · DO

Story points measure the wrong thing.

What breaks

Agile estimation assumes human effort is the bottleneck. When a specification goes to an AI agent and code lands in minutes, story points become noise — the metric measures AI output volume, not the human judgment that decided what to build.

Why AI worsens it

AI flattens implementation cost to near-zero while leaving specification clarity, testing, and integration unchanged — so estimation now optimises for the wrong stage of the pipeline.

Evidence

Stack Overflow Developer Survey 2025 — 66 % of developers report spending more time fixing "almost right" AI code than they save in generation; trust in AI tools dropped from 70 % to 46 % year over year.

Stack Overflow Developer Survey 2025

DoCoDeGo's answer

Replace story points with the ICS gate. A spec scoring below 60 does not begin composition; the focus shifts from "how hard is implementation?" to "how clear is the intent?"

02 Pillar · DE

Sprint cadence breaks when deployment decouples from development.

What breaks

Two-week sprints assume implementation is the bottleneck and shipping on a schedule reduces friction. When AI ships code in hours but human validation still takes days, the sprint clock starts manufacturing artificial delay instead of structure.

Why AI worsens it

Implementation speed now exceeds validation speed; shipping on Tuesday because the sprint ends is no longer defensible when you could ship Wednesday on a passing test.

Evidence

Faros AI 2025 — high-AI-adoption teams complete 21 % more tasks and 98 % more pull requests while PR review time grows 91 %. Overall productivity gain is roughly flat; the bottleneck just moved.

Faros AI 2025

DoCoDeGo's answer

Validation-gated delivery: SDL (Spec-to-Delivery Latency) trending down. Every passing acceptance criterion triggers release readiness; no calendar-driven delay.

03 Pillar · GO

Retrospectives become unfalsifiable.

What breaks

Sprint retrospectives ask "what could we do better?" and rely on the team's honest self-assessment. With AI in the loop, that assessment becomes a feelings report — and the feelings are unreliable.

Why AI worsens it

AI amplifies whatever signal you already optimise for; if you measure output velocity, the velocity chart goes up while real productivity drops — and no one can tell from the metric alone.

Evidence

METR randomised controlled trial, July 2025 — 16 experienced open-source developers predicted AI would speed them up by 24 %, reported feeling 20 % faster, and were actually 19 % slower on real-world tasks.

METR RCT, July 2025

DoCoDeGo's answer

Replace self-report with measurement: AAR (Agent Alignment Rate) tracks how often AI output passes acceptance criteria on first attempt. Retrospectives become "what should our specs say differently?" — driven by data, not intuition.

04 Pillar · DO

PI planning becomes ceremony for ceremony's sake.

What breaks

Scaled-Agile portfolio hierarchy (PI Planning, ARTs, story-point capacity projections across many teams) was designed to synchronise human implementation work across an organisation. When each team's AI ships independently and asynchronously, the synchronisation calendar becomes overhead.

Why AI worsens it

Synchronisation overhead is justified by the cost of mis-coordination; AI removes most of that cost between teams whose work can ship independently.

Evidence

Practitioner reports across 2025 — teams running PI Planning with AI-assisted delivery describe the planning event as "estimating a number that has lost its meaning" and "committing to capacity in a unit we no longer measure honestly".

DoCoDeGo's answer

Replace Epic → Feature → Story with the spec hierarchy: Product Context → Domain Spec → Component Spec. Specs are commands to agents, not estimation containers. "Capacity" becomes "which specs reach ICS ≥ 60".

05 Pillar · DO

"Definition of Done" becomes ambiguous.

What breaks

Agile's Definition of Done is usually "code reviewed, tests passing, deployed". Under AI, that says nothing about whether the deployed system matches what was intended — only that something runs and the tests we already wrote happen to pass.

Why AI worsens it

Implicit assumptions a human author would have carried are invisible to an AI agent and to the DoD checklist that assumed those assumptions.

Evidence

Veracode 2025 GenAI Code Security Report — AI-generated code contains security flaws on 45–88 % of evaluated tasks (Java >70 %, XSS 86 %, log injection 88 %); the rate is not improving as models grow larger.

Veracode 2025 GenAI Code Security Report

DoCoDeGo's answer

The specification is the Definition of Done. Outputs pass acceptance criteria defined in the spec, including a threat model. Statute 15 (Security by Design) makes the threat model non-optional.

06 Pillar · CO

Role confusion: is the developer a coder or an AI supervisor?

What breaks

Agile defines the developer as implementer. AI shifts the day-to-day to a mix of spec-writing, agent orchestration, output review, and architectural defence — four different skills, none of which the existing role description names.

Why AI worsens it

When a role description does not match the actual work, time leaks into ambiguity — and the slowest people are often the most senior, because they hold the most accountability for outputs they no longer wrote.

Evidence

The METR slowdown is partly cognitive overhead — developers reported context-switching between "writing code", "directing AI", and "auditing AI" with no clear hand-off point. Industry surveys in 2025 consistently flag role-clarity as the top adoption blocker.

METR RCT, July 2025 · ThoughtWorks Technology Radar 2025

DoCoDeGo's answer

Four explicit human roles: Intent Architect (DO), Composition Lead (CO), Flow Steward (DE), Governor (GO). One accountability per pillar. In small teams one person holds several; in large teams the roles specialise. No new headcount required.

07 Pillar · DO

"Working software over documentation" is now dangerous.

What breaks

The original Agile value made sense when documentation was overhead and humans wrote the code. Under AI, working software can be produced from a vague spec in minutes — and "working" no longer implies "matching intent".

Why AI worsens it

AI produces output that feels right; subjective evaluation passes; the gap between "the system works" and "the system does what we meant" widens silently.

Evidence

ThoughtWorks Technology Radar 2025 named spec-driven development "one of the most important practices to emerge in 2025" — a direct response to teams shipping plausible-looking AI output that did not match what they actually wanted.

DoCoDeGo's answer

Intent over Implementation — the first DoCoDeGo value. Specifications are the primary artefact; working software is a compiled derivative. A spec without working software is defensible; working software without a spec is not.

08 Pillar · GO

Velocity becomes metrics theatre.

What breaks

Velocity charts go up because more code ships; leadership reads the chart as productivity; the team knows something is wrong but cannot articulate it because the metric does not measure what matters.

Why AI worsens it

Metrics designed for human-driven output measure quantity; AI can produce quantity; only humans can verify outcome — so a metric that worked for humans now lies under AI.

Evidence

Faros + METR composite — output metrics rise sharply with AI adoption while outcome metrics (validated value delivered, alignment-with-intent) move flat or down.

Faros AI 2025 · METR RCT, July 2025

DoCoDeGo's answer

Replace velocity with SDL + AAR + GTR + DDL — four metrics covering delivery latency, alignment, governance trigger frequency, and drift detection latency. None of them lie under AI.

The Agile shift

Same values. Different bottleneck.

The Agile Manifesto's values still hold. They inherit cleanly into DoCoDeGo — what shifts is where the constraint binds.

Agile (2001)
DoCoDeGo (2026)
Why the shift
Working software over documentation
DO Intent over implementation
Specs are now the command interface for AI agents
Responding to change over planning
DE Flow over releases
AI collapses build time; the constraint is validation
Individuals and interactions over processes
CO Direction over production
AI agents are participants, not tools
Customer collaboration over contract negotiation
GO Governance over process
Active alignment replaces passive compliance
The closing argument

We are not abandoning Agile.
We are extending it.

Teams practising Agile well — disciplined, intent-driven, delivery-focused — are already close to DoCoDeGo. The migration is a reorientation, not a replacement. The values you fought for stay. The ceremonies that no longer serve the bottleneck retire.

Read the Manifesto → Explore the Four Pillars →
Alpha · Honest about it

The framework is real.
The community is forming now.

DoCoDeGo is in Alpha. The framework is documented, the practices are battle-tested at small scale, and the next release is being shaped in public.

If it produces anything, it should produce engineers and teams who think more clearly about what they are building and why.

Two doors in
Join the Discord

Discord is where specs are debated, the framework gets sharper, and decisions land in writing. The conversation is the artefact.