“We bought AI coding tools, but we can’t show they improved delivery.”
Last verified: October 2026· engagement
Licences are bought, usage dashboards look busy, and nobody can say whether delivery got faster or quality got worse. The fix is rarely more tooling. It is a baseline for one team, a golden path that team actually uses, and measures that show whether it helped — before anyone rolls it out further.
Is it worth fixing?
It usually is when several of these are true:
- You are paying for seats across a team large enough that a real effect would matter, and leadership is asking for evidence.
- Adoption is uneven: some engineers swear by the tools, others have stopped using them, and review load is creeping up.
- The codebase has properties that make agents thrash — weak tests, fuzzy boundaries, undocumented conventions.
- A renewal or wider rollout decision is coming and you have nothing measured to base it on.
The decision, one question at a time
Each step links to the page that answers it. You can get most of the way without talking to anyone.
- 1
- 2
- 3
- 4
- 5
Get it approved
What does a pilot need and what counts as done?
The workflow engagement, terms included →
What a first engagement looks like
Free read
A written read on the top three gaps holding your rollout back, from how your team uses the tools today.
Paid diagnostic
A fixed-price baseline of one team’s delivery measures plus a codebase-readiness read, with a go/no-go on a pilot. Fixed price, quoted after the free read.
Stop or go
If the baseline shows the constraint is not tooling — process, ownership, or the codebase itself — we say so before any build.
The accepted outcome is the test above: improvement in the delivery measures chosen at kickoff for one pilot team, without worse defect rates or review burden.The full acceptance criteria, exclusions and pricing terms are written out onthe engagement page, so you and procurement can see the finish line before any call.
Access and security
- Read access to the pilot team’s repos and CI.
- Delivery data you already collect; we do not need individual-level tracking.
- Approval from whoever owns tooling and security policy.
When we’d tell you not to buy
Our scarce resource is senior engineering attention, so we would rather say no early. We decline or refer when:
- The goal is individual productivity scoring or headcount justification for specific people.
- There is no team willing to pilot and no owner to run it.
- You want an org-wide rollout in the first engagement with no pilot.
Not there yet? Start withHow to measure AI coding ROI, The AI coding golden-path playbook or Context engineering for agents.
The engagements behind it
Internal AI Coding Workflow Build-Out
$40–120k build, $5–10k/mo tuning · 6–10 weeks
Agentic Codebase Readiness Audit
$10–25k audit (often leads to remediation) · 2–3 weeks
Questions
Why start with one team?
Because a baseline and a golden path that one team actually uses tell you more than an org-wide rollout nobody can measure. The second team is cheaper once the first works.
Do you measure individual engineers?
No. We measure team-level delivery, defects and review burden. Individual productivity scoring is not something we do.
What if the problem is the codebase, not the tools?
That is a common finding. The baseline includes a readiness read, and if the constraint is tests, boundaries or conventions, we tell you that before proposing any tooling work.
Related guides
What is agentic coding?
Agentic coding is letting an AI coding agent plan, edit across files, run tools and tests, and iterate in a loop toward a goal — powerful for the first 80% of a change, and quietly expensive on the last 20% if you skip standards and review.
Agentic codingHow to measure whether AI coding tools are working
The metrics vendors hand you — seats active, suggestions accepted, lines generated — measure tool usage, not delivery. The honest version starts from the delivery metrics you already track, adds two or three that are specific to agent work, and accepts that the result will be a directional read rather than a controlled experiment.
Agentic codingContext engineering for coding agents
Context engineering is the work of deciding what a model sees on each turn: what goes in, in what order, and — the part teams skip — what gets left out. On coding agents it is usually the difference between an agent that lands a clean diff and one that thrashes, because almost every agent failure that looks like a reasoning failure is really a context failure.
Agentic codingIs your codebase ready for AI agents?
Agents don't fail randomly on a repo — they fail predictably on the same things that slow down human engineers: fuzzy boundaries, missing tests, undocumented conventions. Readiness is just those weaknesses paid down deliberately.
Find out whether it’s worth fixing
A few structured questions, no call required. You get a written decision brief back — including whether we think you should buy anything at all.
Start the diagnostic