AI writes the code. Your team pays the review tax.
Output metrics climb while the real work moves to review, judgment, and rework — and no delivery dashboard bills you for it. Forge reads that cost straight from your merged pull requests, then sizes every commitment to the capacity your team actually has.
Private repo? npx @ambera/review-tax runs locally, sends nothing.
A layer on the stack you already run — never a migration
The review tax
Output is up. This is what it’s costing.
Merged PRs and story points keep climbing — those are the metrics dashboards celebrate. Underneath, the human work has moved to review rounds, rework, and judgment on code nobody wrote by hand.
- 01
Forge pairs each merged PR to the work it came from and separates the ones carrying agent authorship from the rest.
- 02
It compares what each cost in review rounds, rework pushes, and time to merge — per area, over time.
- 03
The attributed share is a floor, never a verdict — tools that only complete code inline leave no trail, so “rest” is not the same as human-written.
cli/cli · 100 most recently merged PRs
≥54% attributedOn GitHub’s own CLI, pull requests carrying agent authorship drew 3.6× the reviews of the rest — and took 2.1× as long to merge.
36 of its last 100 merged pull requests ran attributed-heavy. The output chart shows a faster team; the review loop underneath is carrying the difference.
See the live report ↗ — it re-reads over time, so today’s numbers may differ.
0.0×
the reviews submitted of everything else
≥0%
of merged commits carrying agent attribution — a floor
0hvs87h
time to merge, attributed vs rest
From the live scan of cli/cli, read Aug 2026. Across the public directory, roughly half of repos pay a premium and half don’t — the point is knowing which yours is. Run your own in seconds.
Estimate → actual
An estimate is a range. The actual comes back to meet it.
Most trackers take a number, then never mention it again. Forge keeps the band, then tells you where the work actually landed inside it.
A band, not a number
Realistic is the energy-aware estimate; optimistic and pessimistic span the confidence band. The pessimistic side leans wider, because work overruns its plan more often than it beats best-case.
Width learned from this person
The band tightens as someone’s estimates land close, and widens when they don’t — sized from their own history, not a team-wide fudge factor.
The actual comes from the merge
Nobody fills in a form. When the pull request lands, Forge reads what the work actually cost from the review loop and grades the estimate against its band.
Planning on real capacity
Everything your delivery dashboard hides.
Velocity charts average away the person, the calendar, and the review load. Forge plans on what’s actually underneath.
Estimates sized to whoever picks the work up
A senior who owns the service and someone in their first week don’t carry the same cost. The estimate — and its confidence band — follows the person, learned from how they’ve actually closed work.
A QA reserve, held on day one
Every sprint keeps a visible slice for testing, review, and the unplanned — not discovered on the last day.
Over-commitment, caught with runway
A plan running past sustainable pace gets flagged on day two — not in the retro.
Capacity reads the real calendar
Holidays, booked time off, part-time weeks — the plan starts from the days your team actually has.
Review concentration, never a ranking
See when review load piles onto too few shoulders — as a distribution, no per-person leaderboard. A published promise.
A weekly brief that reads the week for you
What last week revealed, which of this week’s commitments look expensive, where the reserve is thinnest.
Meets your team inside the PR thread
A comment bot pairs pull requests to their tasks as they merge, and the MCP connector puts Forge in your agent’s hands.
How it fits
A layer, not a migration.
Nothing to move, no process to change, nothing new for the team to adopt. Forge has a full planning surface of its own if you ever want one — that’s an option, never the price of entry.
Connect a repo
Read-only, merged pull requests only. Keep planning in Linear, Jira, GitHub Issues, or ClickUp — Forge reads alongside the tracker, it doesn’t replace it.
Forge reads the review loop
Every merged PR is paired to the work it came from and graded on what it actually cost — review rounds, rework pushes, time in review — attributed vs the rest.
Plans sized to real capacity
Estimates follow the person, the calendar subtracts holidays and time off, a QA reserve is held up front, and over-commitment is flagged while there’s runway.
The privacy membrane
Your energy stays yours.
Forge reads the human cost behind the work without ever handing anyone’s personal data to the org. Each person’s ledger lives in their own account; the organization sees the plan — never the reason behind it.
Private to each person
- The energy ledger and how a day felt
- Why an estimate moved up or down
- Recovery, depletion, and check-in history
What the org ever sees
- De-identified story-point estimates
- Confidence bands on each estimate
- Sprint capacity and reserve, in aggregate
Enforced structurally, not promised in a policy — personal data never crosses to the org’s side, so there’s nothing there to leak or subpoena. The promise is published, from the engineer’s side, at /privacy-for-engineers.
Proof over promises
Judge it before you trust it.
No borrowed benchmarks, no confidence you can’t see, no data leaving your machine until you decide. The trust is built into how it works.
Every repo anyone has read is browsable in the public reports directory — yours might already be in there.
Pricing
You pay for who ships. Not for who logs in.
Viewers are free — invite the whole company. What you pay for is coverage: the engineers whose pull requests Forge actually reads, counted from the repo rather than a seat list somebody maintains.
Forge Free
$0forever
Up to 10 contributors
Get the number. No account for public repos.
- —Unlimited public-repo reads — no account, nothing installed
- —Your own private repos, refreshed monthly
- —Shareable report, README badge, and embed
- —Cohort benchmark and repo-to-repo compare
- —npx @ambera/review-tax — read any private repo locally, sends nothing anywhere
Forge Team
$299/ month
15 contributors included, then $12 each
The review tax on your own code, continuously — read by everyone.
- —Unlimited private repos, re-read on every merge
- —Unlimited viewers — you pay for who ships, never for who looks
- —Attributed vs unattributed pull requests, per area, over time
- —Per-engineer story-point estimates with confidence bands
- —Actual effort graded from the review loop — rounds, rework, time in review
- —Commitments sized to real capacity, with over-commitment caught early
- —Linear, Jira, GitHub and calendar sync — keep the tracker you have
- —Weekly brief, PR comment bot, and AI assists on your calibration history
Forge Org
Customfrom 60 contributors
Annual invoicing, DPA, security review
Every team in one read, for the person who owns the quarter.
- —Everything in Team, across every team, with cross-team rollups
- —Where AI helps and where it costs you, mapped per area
- —Comprehension debt — code shipping faster than anyone is absorbing it
- —Workload watch across the org, with the privacy membrane intact
- —Benchmark against your industry cohort, not the public average
- —SSO, audit trail, and methodology export for board reporting
A covered contributor authored or reviewed a pull request in the last 30 days. Bots don’t count, quiet months don’t count — the number is measured, never a seat bought in advance. Personal energy data stays private on every plan; the workspace only ever sees the plan.
Ambera, for you
The same instrument, at personal scale.
Forge reads the cost of work at team scale. Ambera is where the idea started — a debt tracker for your own energy. Log what spends and restores you, watch the running balance, and notice the imbalance before you feel it.
- Today’s net and the rolling 7-day balance
- The cumulative debt forecast, honest about stale data
- A quiet nudge when the week drifts depletive
The founding story
Every company I worked at ran on the same quiet assumption — that a good engineer can always give a little more. There’s always one more sprint, one more launch, one more thing only you can ship. And the people who keep answering that call, the ones who care most about the work, are usually the ones who pay for it.
I was one of them. I pushed hard everywhere I went, because the work mattered and because that’s what good engineers do. And every time, the same thing happened — the fire went out before I noticed it was burning low. Not because anyone was careless, but because nothing was keeping score. My output was tracked to the hour; what it cost me was tracked by no one.
That’s the part that stayed with me. The burnout was never the surprise. The surprise was that it arrived in silence — by the time I could feel it, the debt had been building for weeks. Feeling it was the last step, not the first.
An ember is heat held steady. A flame burns bright and consumes its fuel — the engineer who ships hard for eighteen months and then can’t open their laptop. A cold coal is the other way to lose the fire. Staying at ember temperature is a deliberate act, and it takes an instrument to do it.
Ambera is that instrument — first for me, now for you, and with Forge, for the teams whose dashboards still only track the output. A thermostat for effort, not a thermometer for mood.
— Maki, founder
Start free
Start with the number, not the signup.
Pick any public repository and see its review tax in seconds. If the number looks familiar, Forge reads it continuously on your own code.
Private repo? npx @ambera/review-tax — runs locally, sends nothing.