→Navigara · Product·Engineering Health

Engineering process
health.

Find the stage that holds your delivery back: coding, code review, or product context. Check engineering process health every night.

Trusted by engineering teams

Kiwi.comFinshapePartners BankaPurple TechnologyGreysonItrinityESETFTMOKeggSecond Foundation

00 · The loop

Engineering process health across coding, review and product.

01

Coding

Speed it up and review becomes the constraint

100%
02

Reviewing

Speed both up and product feedback becomes the constraint

62%
03

Product

When specs and tickets stall, coding is starved again

24%

Whatever leaks at the narrowest stage is the constraint. It feeds straight back into coding.

01Part 1 · Coding

Every developer knows their next step with AI.

The AI adoption framework shows each developer where they already work well with AI and the one next step that unlocks more. Built on evidence from the work itself, not on self-reports.

AI adoption framework

Five levels, each scored from its own evidence and read separately. There is no combined score, because the lowest row is the one worth working on.

Jun 8 to Aug 2 · 8 wks · n=42
  • Agent fluency4.4 of 54.4↑ +0.5Strong habit. Keep AI in the daily loop, 7 of 8 weeks active
  • Delegation depth3.9 of 53.9↑ +0.9Hand agents whole tasks, not snippets, on every agent day
  • Roadmap alignment3.5 of 53.5→ steadyGrowing. Point more AI hours at roadmap objectives
  • Context leverage0.8 of 50.8→ steadyBiggest opportunity. Add a why and outcome to every prompt and ticket
  • Agentic autonomy0.5 of 50.5↑ +0.4Ready for more. One agent-owned task a week, end-to-end

Context leverage is the biggest opportunity on this team, at 0.8 of 5 while the rest of the profile is fluent. Every developer gets the same read-out for their own work, so the next step is theirs and not a team-wide mandate.

02Part 2 · Reviewing

Find code review bottlenecks.

When agents write code faster than people can review it, the review queue grows. Engineering process health shows code review bottlenecks through change size, wait time, and review activity. A review agent posts a first pass; your team still owns review, tests, and the merge.

Time to first review4.2h↓ −11.3h vs Q1
Time to rework6.5h↓ −2.1h vs Q1
Merged w/o review6%↑ +2pts, flagged weekly
Agent first pass38%of PRs over 400 lines

When a change is too big for the queue, the review agent posts a full first pass: risk map, test gaps, and the three files a human must read.

03Part 3 · Product

Improve the context in objectives and tickets.

Fast coding against a vague objective is expensive guesswork. Navigara audits every objective, epic and ticket for context health: how much of its context a reader can actually find.

Overall context health
29%across 16 objectives
  • 2healthy
  • 1thin
  • 13poor

13 of 16 objectives cannot answer basic questions about why they exist.

Stable 1.0 APIs and ComponentsActive15%
  • Problem, risk, or opportunityMissing0/15
  • Customer or affected partyMissing0/15
  • Expected value and reason for the betMissing0/15
  • Success measure or evaluation planMissing0/10
  • Delivery approach and sub-objectivesIn the record10/10
  • Current status and next reviewIn the record5/5

Next step: add problem, risk, or opportunity to the description. 8 of 10 questions have no answer, and the 15 points earned are out of 100.

Context health is scored per objective, and it rolls up to the roadmap so you can see which priorities are being built on a record nobody can read.

See roadmap alignment
04The agent

Healthy processes are the key to faster delivery.

Nightly checks link every finding to the pull request, ticket, or sprint behind it. Engineering process health then ranks code review bottlenecks and context gaps by severity, and your team picks the fix.

NProcess checks · last nightly runNavigara agent1 passing · 5 need attention
Open findings124
High severity4
Resolved78
  • Rubber-stamp review on a high-risk changevision · PR #20044 open21 fixed
  • Work shipped under an epic the roadmap never trackedENG-288466 open12 fixed
  • Team sprint spent on a deprioritized initiativePlatform · sprint 419 open4 fixed
  • Commit unrelated to the ticket it referencesvision · PR #211814 open3 fixed

The highest-severity findings from 4 of 6 checks. Each one opens onto the pull request, ticket or sprint it was raised against, so the fix is a click away rather than an investigation.

The full check library, what each one reads, and what a finding looks like when it lands.

See Agent Speed
05The cost of AI

What did all this AI cost, and which of it did anyone ask for?

This illustrative report shows $226,900 of AI spend over 30 days, split by work type and roadmap alignment. Review maintenance and fixes in context before assuming they are waste. Work without a stated outcome may need clearer tracking or a priority decision, rather than an automatic budget cut.

NAI spend · last 30 days$226,900
By work typeWhat were the tokens for?
Features 51.8%$117,534
Maintenance 22.3%$50,599
11.9%
5.1%
8.9%
Features $117,534Maintenance $50,599Tests $27,001Docs $11,572Fixes $20,194
By roadmap alignmentDid anyone ask for it?
Roadmap aligned 48.6%$110,274
Aligned 22.7%$51,506
Unaligned 28.7%$65,120
Roadmap aligned $110,274· tied to a named objectiveAligned $51,506· justified, off roadmapUnaligned $65,120· no stated outcome
Unaligned · no stated outcome$5,290

Fix the specs in part 3 and the unaligned block shrinks. Route each task to the cheapest model that can do it and the whole bar does.

See token spend intelligence
06Monday morning

Act on code review bottlenecks and missing context.

  • Name the stage that is actually your constraint, instead of speeding up the one you can already see

  • Give every developer one next step with AI, scored from their work and not from a survey

  • Investigate code review bottlenecks before a team starts clearing the queue by skipping review

  • Stop funding work that has no stated outcome behind it

Engineering process health questions.

What does engineering process health cover?

Engineering process health covers three stages: coding, review, and product context. The narrowest one sets your pace. Code review bottlenecks often appear once agents speed up coding, and a vague objective can slow all three.

How do we find code review bottlenecks?

Look at change size, time to first review, and review activity, team by team. Engineering process health links each finding to the pull request behind it, so a four-hour wait comes with its cause.

Does engineering process health replace human review?

No. Your team owns review, testing, and merging, and engineering process health keeps it that way. A review agent can post a first pass on large changes, and code review bottlenecks get fixed without dropping the checks that catch errors.

Why do objectives matter to engineering process health?

Fast coding against an unclear objective produces work nobody needed, quickly. Engineering process health scores each objective on whether it states the problem, the customer, and the expected outcome. Fix those gaps alongside code review bottlenecks, or faster flow ships the wrong thing sooner.

How do AI adoption and code review bottlenecks interact?

More coding throughput can move the constraint into review and create code review bottlenecks. Engineering process health reads the whole loop, from adoption to product context, so you can see where the queue moved after agents arrived.

Where should we start with engineering process health?

Connect your repositories and tracker (read-only), and open the highest-severity engineering process health finding. Make one change with the team. If code review bottlenecks were the issue, the next run shows whether waiting times dropped.

→Start here

You can’t improve
what you can’t see.

Inspect code review bottlenecks and product context on your own repositories. Setup is read-only, and the first run uses the history already in scope.

Coding

AI adoption, measured on evidence

Reviewing

Code review bottlenecks, with the evidence behind them

Product

Healthy specs before work starts

How the checks run