Skip to content

Your whole team has Claude. Is it paying off?

Promptster reads how your team actually works in the AI coding tools they already use, then shows where the tooling pays off, where spend leaks, and gives each engineer private, targeted coaching to close the gap.

Book a 15-min walkthrough See a sample dashboard
Overview
ManagerDeveloper
Platform Engineering· 44 eng
This weekFull
Last 7 daysCustomizeExport
This month · outcome

What the platform put back

Net saved / mo
$33.6k
8.0× return on spend
Platform spend / mo
$4,820
$127 / eng · +11%
Waste recovered / mo
$3,040
▲ vs last month
Skill-tooling spend / mo
$215
5 unhealthy · $47 burning
Delivery · DORA + GitHub

Did it move shipping?

Connected sources
GitHub
342 PRs · synced 4m
Jira
88 issues · synced 11m
PagerDuty
24 incidents · synced 6m
Deploy freq
9.4 /day
Elite · +1.2/day
Lead time
21 hrs
−4.5 hrs faster
Change-fail
4.1 %
+0.3 pts · watch
MTTR
1.8 hrs
−22 min
Delivery up since AI rollout
vs Q3 '25 pre-AI baseline
Merge rate+14%
Cycle time−22%
Rework−9%
This week · actionable

Recommend a habit to the whole team

Prioritized from your telemetry. Roll one out, Promptster captures the baseline and proves the lift in dollars and code quality.

Run a verification-loop clinictier lift
38% of the team sits at Developing on the weakest dimension this month.
Roll out migration-runner team-wide+36% reach
Grade A · 94% success · but 41% reach, Identity squad only.
Share the context-hygiene playbook~$1,100/mo
Infra squad's context efficiency trails the team by 18 pts.
You rolled this out last week
Plan-mode clinic · whole team
Saved / wk
+$2.4k
Rework rate
−4.2pt
Plan-mode adoption47% → 58%+11
Verification before done54% → 61%+7
Attributed lift · same cohort · insight → action → moved.
Integrations · no tool mandates

Works where your engineers already work.

Codex
  • task briefsthe spec they hand the agent
  • transcriptscourse-correction when it drifts
  • edit cadencewhat ships vs what gets reworked
  • test runsproof it works, or a guess

measured: spend · context · workflow · skills

Claude Code
  • plans & promptsdo they scope before coding
  • agent transcriptssteering vs rubber-stamping
  • shell commandshow they recover when stuck
  • edit cadenceaccepted as-is vs reworked

measured: spend · context · workflow · skills

Cursor
  • agent chatsthe context they feed it
  • inline editssurgical fixes vs full-file pastes
  • terminalhow they debug when it breaks
  • test runsverified or vibes

measured: spend · context · workflow · skills

Copilot
  • chat & promptsthe ask behind each suggestion
  • completionsaccepted as-is vs reshaped
  • edit cadencewhat ships vs what gets reworked
  • test runsproof it works, or a guess

measured: spend · context · workflow · skills

Claude Code, Codex, Cursor, and GitHub Copilot are all fully supported.

The problem · self-report is broken

Engineers can't tell you how good they are with AI.
The research shows why.

The cost isn't the subscription. It's the salary next to it.

One engineer$200,000 / yr
output you're paying for19% drag from unmeasured AI use ≈ $38k / yr: the slowdown METR measured in devs who felt faster
The AI seat next to that bar costs $2,400 / yr, about 1%. The risk was never the subscription.
−$38k / yrthe drag METR measured, for every engineer who feels faster but isn't, until someone can see it
5× possiblethe productivity vendors' own case studies report: the ceiling your team is paying for but not collecting

Every chart above is self-report or an aggregate. None can name who on your team uses AI well.

How it works · always on, no exam

Set it up once. It runs on the real work.
No homework, no two-week clock.

01

Connect.

One CLI instruments the AI tools your engineers already use (Claude Code, Codex, Cursor, Copilot), scoped to the repos you choose.

# platform team · one-time setup
❯ promptster connect
✓ 3 repos · open-source client · 23 engineers notified
02

Capture, continuously.

Real work sessions as they happen: no exam, no simulation day, no clock. Promptster reads where spend leaks, where context thrashes, and where the workflow could be tighter.

● live · always on
spend · context · workflow · skills
23 engineers · 3 repos · real work
03

Act.

Managers get an aggregate team dashboard: savings, fluency, skill adoption. Each engineer gets a private view with targeted, self-directed fixes.

# two surfaces, one wall
manager → team trends
engineer → private coaching
The product · two views, one wall

Managers see the team.
Engineers see themselves, and only themselves.

Same captured work, two views. Individual numbers stay private to the engineer. Architecture, not policy.

Team-level only · no names, no per-engineer ranking.

Overview
Platform Engineering· 44 eng
This weekQuarter
Last 90 days
Aggregate · this quarter

Team AI-fluency · tier mix

44 engineers · all 5 team dimensions
68th percentile · +12 pts/qtr
Strong · 15Adequate · 21Developing · 8+7 moved out of Developing this quarter
Direction quality↑
direction_quality
Steering discernment↑
steering_discernment
Verification loopNext focus↑
verification_loop
Context management↑
context_management
Ecosystem leverage→
ecosystem_leverage
Adoption · last 7 days

Who's actually using it

Active AI users
38 / 44
+4 this week
AI-assisted PRs
71%
+6 pts
Team fluency
68
▲ +12 / qtr
Aggregate only. No engineer's per-dimension tiers appear here — each sees their own privately.
The rubric is yours

Every team starts on our open-source base rubric: dimensions anchored on Anthropic's AI Fluency, kept at the frontier as the agents evolve. If your team leans on subagents, ships test-first, or has its own house style, we tune the dimensions and anchors to reward how you actually work. You see every dimension; the calibration underneath is ours.

Read the rubric on GitHub
Open the full-screen demo
Always current · staying at the frontier is our job

The best way to use these tools
changes faster than any team can track.

Claude Code ships new features weekly. The models change under you. The patterns the best engineers lean on today didn't exist last quarter. Keeping a whole team current on all of it is a full-time job — so we make it ours.

Your telemetry tells us how your team works. The frontier tells us how the best teams work right now. Promptster folds both into the recommendations every engineer and manager sees — so the guidance is never generic, and never stale.

  • We track the frontier, not you. New agent features, prompting patterns, and workflow shifts — we watch the whole industry and distill what actually moves output.
  • Best practice, delivered as a nudge. When the state of the art moves, it arrives as a concrete, prioritized recommendation in-product — not a newsletter your team won't read.
  • The rubric stays at the frontier too. The dimensions we score against evolve as the agents evolve, so “good” always means good today — not what was good six months ago.
Why not what you already have

Nothing in your stack
watches the actual work.

Surveys, AI upskilling platforms, and DevEx dashboards all orbit the question and miss it completely. None of them can tell you where the leverage and the money actually leak, or how to fix it, because none of them see the work itself.

Comparison of surveys, quizzes and certifications, DevEx dashboards, and Promptster across four dimensions of AI-fluency measurement.
DimensionSurveys / self-reportWhat you haveAI upskilling platformsWhat you haveDevEx dashboardsWhat you havePromptsterObserved sessions
What it measuresPerception.General AI literacy, not engineers in real codebases.Aggregate output: DORA, throughput, cycle time.Observed behavior in real sessions: spend, context, workflow, skills.
Tells you WHAT to changeNo, just a mood.A generic course.No, a number with no fix attached.Yes: the exact model, habit, or skill, from real moments.
Respects engineer privacyAnonymized, so unactionable.Scores held over people.Aggregate only.Individual detail stays with the engineer, never the manager.
What it measures
Surveys / self-report
Perception.
AI upskilling platforms
General AI literacy, not engineers in real codebases.
DevEx dashboards
Aggregate output: DORA, throughput, cycle time.
Promptster
Observed behavior in real sessions: spend, context, workflow, skills.
Tells you WHAT to change
Surveys / self-report
No, just a mood.
AI upskilling platforms
A generic course.
DevEx dashboards
No, a number with no fix attached.
Promptster
Yes: the exact model, habit, or skill, from real moments.
Respects engineer privacy
Surveys / self-report
Anonymized, so unactionable.
AI upskilling platforms
Scores held over people.
DevEx dashboards
Aggregate only.
Promptster
Individual detail stays with the engineer, never the manager.
Security & compliance · built for engineers

Built for engineers,
not a manager's scoreboard.

An individual engineer's numbers stay with that engineer, never the manager. Managers see team-level trends only: no per-person scorecard, no ranking, enforced by RBAC, not a policy promise. On top of that: read-only, source-excluded, and built on the controls your security team already asks for.

  • Private to the engineer. Individual metrics and coaching go to that engineer, and only that engineer. Developmental, never a review input.
  • Aggregate to managers. Managers see team-level trends only: no individual scorecard, no ranking, by RBAC not policy.
  • Source excluded. We read prompt context, never your source code. Enforced at ingestion, not promised in a policy.
  • SOC 2-equivalent controls. Encryption in transit and at rest, least-privilege access, and audit logging. A formal report is on the roadmap.
  • ≤ 90-day retention. Prompt context expires within 90 days by default. Retention window and data residency tune to your contract.
  • Deletion on request. Delete any time, plus automatic expiry, plus a full purge of an engineer's data when they off-board.
capture.manifest promptster-teams-cli
Captured
  • prompts & plans
  • tool calls
  • terminal commands
  • test runs
Never captured
  • your source code
  • keystrokes
  • screen recording
  • clipboard
  • webcam
  • anything outside scoped repos
The shape of a PR description, never the code itself, captured by a client you can read line for line.
  • Are you SOC 2 certified?
    We run a SOC 2-equivalent control set today: encryption in transit and at rest, scoped least-privilege access, and audit logging. A formal report is on the roadmap, and we're happy to walk your security team through the controls on a call.
  • What do you store, and for how long?
    Prompt context only, never source code, encrypted, with ≤90-day retention by default. Retention window and data residency can be set per contract.
  • Can we delete our data or off-board an engineer?
    Yes. Deletion on request, automatic expiry at your retention window, and a full purge of an engineer's data when they leave the org.
  • Who can see an individual engineer's numbers?
    Only that engineer. Managers get team-level aggregates; RBAC enforces the split by architecture, so a per-person leaderboard isn't something the product can produce.
See it on your own team

Book a 15-min walkthrough.
We'll show the live dashboards.

Bring a VP Eng or platform lead, and we'll show the manager view and an engineer's private view, and you decide in 15 minutes whether to connect a repo and see your own.

Book the walkthrough
or get the monthly memo
no spam · we reply personally