Assistant dashboard
31%
suggestions accepted
Cursor and Copilot now bill by usage
AI already writes part of your code. HubBound shows you which part, what it cost and whether it worked.
HubBound Analytics is AI analytics for engineering teams. It joins every session from Claude Code, Codex, Cursor, GitHub Copilot and Antigravity with your tickets, pull requests, releases and Datadog incidents, so you can see what the AI did, what each ticket cost and whether it reached production without problems.
Illustrative example from Northstar Labs: one ticket, from the AI session to a production signal.
The problem
Your team writes code faster. Does it ship faster?
AI sped up the part that was already fast. PRs pile up in review, the bill changes every month and every tool shows you only its own slice.
AI billing
US$8,120
this month, usage-based
Jira
146
tickets closed
What did that US$8,120 buy?
Illustrative figures for a fictional team.
What the latest research says
vs +30%
With autonomous AI agents, commits rose 240% while releases rose only 30%, across more than 500,000 GitHub developers.
More than 80% of nearly 5,000 professionals believe AI raised their productivity. Even so, AI adoption still tracks with lower delivery stability.
That is the median speedup the 349 technical professionals METR surveyed between February and April 2026 (published in May) say AI gives them. METR itself warns it is perception: its latest experiment measured 4% to 18% less time, with weak evidence.
This is what a ticket looks like when you follow it end to end.
TCK-184 went through three AI sessions, one pull request, one release and seven days in production without incidents. HubBound puts those pieces together from the tools you already use, wherever tickets, commits and identities can be linked. And what never reaches a ticket shows too: AI sessions without a ticket appear with their time and cost.
AI sessions
Who used which tool and model, what each session did, how many tokens and what it cost.
TCK-184: 3 sessions in Claude Code and Cursor
Claude Code
Codex
Cursor
GitHub Copilot
Antigravity
Tickets and code
Which ticket and PR each session went to, how long review waited and whether it merged.
TCK-184: PR #412, first review in 1.8 h
Jira
Linear
GitHub
Production
When it reached production and whether it needed a rollback or a hotfix.
TCK-184: REL-042 in production in 7.9 h; checkout-api with no incidents in 7 days
Datadog
Team experience
Whether the team feels AI helps or adds review work, answered anonymously in Pulse Survey, with a response minimum that protects whoever answers.
Team: Read per team
- Pulse Survey
Now multiply that by 146 tickets and 48 people.
Open any AI session and understand what happened.
Every session comes with a summary, the time actually worked, the cost, the model and the ticket it was linked to. The team's habits come from there: when people work with AI, on what, how sessions end and how well context is handled. It tells you where AI-assisted work gets stuck and which habits to copy between teams.
Illustrative example of an AI session, the team's habits and prompt quality, from Northstar Labs.
Questions it answers
- What happened in this session?
- What does the team work on with AI?
- How do sessions end?
- Is context handled well?
The questions you'll get asked about AI, answered in one demo.
Northstar Labs is a fictional 48-person team with 30 days of illustrative data. Each chapter answers a question that usually comes from finance or leadership.
Where should we act first this month?
−30.8%time to first review
All 4 areas are healthy. Time from commit to production fell 29.5% and review is what improved most this month.
- 412 merged PRs
- 146 tickets
- 36 active people per week
What happens to the code after AI writes it?
82%of AI sessions with commits reach merged code
AI took part in 70% of the 412 PRs, and only 4.7% of changes failed in production, 3.1 points fewer than the previous period.
- 1,478 commits
- 64 deployments
- CI at 96% success
Who really uses AI, and what does it cost?
36 of 48people use AI every week
They use it on 81% of working days. 4 seats sit unused and the month cost US$8,120 across 1,240 sessions.
- WAU 36
- MAU 44
- 4 unused seats
What does AI cost us per ticket?
US$38.40median AI cost per ticket
The average rises to US$55.36 because a few tickets concentrate the spend. Each ticket needed 2.8 PRs.
- 132 tickets with cost
- US$26.31 per PR
- 6% churn
Is going faster costing us quality?
7.9 hfrom commit to production, 29.5% less
The failure rate fell to 4.7% over the same period, so speed was not paid for with quality. AI returns 2.1 net hours per person per week.
- HUBPR 84/100
- flow efficiency 58%
- failures 4.7%
What does the team do when it works with AI?
64%of sessions end completed
Software engineering is 78% of the AI work, and the peak hour runs from 10 to 11. Context hygiene averages 72 out of 100.
- 312 analyzed sessions
- 21% partial
- 9% blocked
Best viewed on a large screen.
HubBound Analytics
Illustrative sampleThe interactive demo could not load. Try again or request a demo for your team.
Illustrative data for a fictional team. This demo does not connect your sources.
Want to see it with your tools and your team?
Request a demoMetrics that hold up when finance asks.
Each one with its definition, a sample value, the question it answers and where it comes from. When someone asks where a number comes from, the answer is right next to it.
Cost
- AI cost per ticket
Cost of the sessions linked to a ticket, calculated with per-model pricing. Median and average are reported.
ExampleUS$38.40median, last 30 daysAnswersWhat does AI cost us for each ticket we close?
Source: AI sessionsSource: Tickets and code- Cost per pull request
Cost of the sessions linked to each merged PR.
ExampleUS$13.40average per merged PRAnswersHow much AI went into each pull request that merged?
Source: AI sessionsSource: Tickets and code- Spend by tool and model
AI spend split by tool and model, with the share each one takes.
ExampleUS$8,120last 30 days, by modelClaude Sonnet 5.5 54%gpt-5-codex 28%gemini-2.5-pro 12%Other 6%AnswersWhich tools and models take the spend?
Source: AI sessions- Spend attributed to delivery
Share of AI spend that can be linked to tickets and pull requests that were delivered.
Example68%of the spend is linked to deliveryAnswersHow much of the spend turned into delivered work?
Source: AI sessionsSource: Tickets and code
Adoption
- Weekly adoption (WAU)
People who used at least one AI tool during the week, over the people eligible to use one.
Example36 of 48people active last weekAnswersHow many people really use AI each week?
Source: AI sessions- Depth of use
Share of working days on which each person used AI (called “frequency” in our adoption guide). It separates habit from access.
AnswersIs AI a habit or just something people have access to?
Source: AI sessions- Unused seats
Paid seats with no activity in the period. Needs the seat source to be connected.
AnswersAre we paying for seats nobody uses?
Source: Seat source- AI fluency
A 0 to 100 score across four dimensions (delegation, description, discernment and diligence) plus habit. It describes how people work with AI.
Example78/100team averageAnswersHow well do people work with AI?
Source: AI sessions
Delivery
- Sessions that reach code
Share of AI sessions with commits that end in merged code.
Example82%of AI sessions with commitsAnswersHow much of what AI writes ends up merged?
Source: AI sessionsSource: Tickets and code- Change lead time
Hours from commit until the change runs in production. A DORA metric.
Example7.9 h29.5% less than the previous periodAnswersHow long does a change take to reach production?
Source: Tickets and codeSource: ProductionHow to measure DORA- Change fail rate
Share of deployments that need immediate intervention (rollback, hotfix or feature flag). A DORA metric.
Example4.7%3.1 points lower than beforeAnswersIs going faster breaking production?
Source: Production- Deployment success
Share of deployments to production that finish well.
Example95%of 64 deploymentsAnswersDo the releases we ship hold up?
Source: Tickets and codeSource: Production- Pull requests per ticket
How many pull requests a ticket needs before it is done.
AnswersDo tickets need many rounds to close?
Source: Tickets and code- Code churn
Share of recently written code that is rewritten soon after. It shows whether AI code holds up.
AnswersDoes the code AI writes last?
Source: AI sessionsSource: Tickets and code
Productivity and experience
- Net hours per person per week
Time AI gives back once the extra review and rework time is subtracted.
Example2.1 hreturned per person, per weekNet 2.1 h 66%Review and rework 1.1 h 34%AnswersHow many hours does AI really return?
Source: AI sessionsSource: Tickets and codeHow to measure AI productivity- Flow efficiency
Share of cycle time in which work is active instead of waiting.
Example41%of cycle time is activeAnswersDoes work flow, or does it wait?
Source: Tickets and code- HUBPR profile
A 0 to 100 profile across five dimensions: AI habits, frictionless flow, solid quality, price-value and reported experience. It draws on DORA and SPACE. Read it next to each dimension.
Example71/100across five dimensionsAnswersWhere is the team strong and where does it need support?
Source: Tickets and codeSource: ProductionSource: Pulse Survey- Pulse Survey responses
How many people answered the team survey. Results are published only above a minimum number of responses.
Example31 of 48people respondedAnswersIs the survey representative enough to trust?
Source: Pulse Survey
Sessions
- Useful sessions
Share of analyzed sessions that produced something worth keeping.
Example74%of 312 analyzed sessionsAnswersIs the time spent in AI sessions paying off?
Source: AI sessions- Context hygiene score
A 0 to 100 score for how cleanly the context window is managed. Higher is better.
Example72/100team average, rated GoodAnswersDo people handle context well?
Source: AI sessions- How sessions end
Split of analyzed sessions into completed, partial, blocked and cancelled.
Example64%completedCompleted 64%Partial 21%Blocked 9%Cancelled 6%AnswersDo sessions finish what they started?
Source: AI sessions- Session length
How long sessions last, as the median and the 90th percentile.
Example38 minmedian; 9 in 10 are under 1 h 50AnswersHow long does the team work in each session?
Source: AI sessions
Ask Pulse AI instead of hunting through a dashboard.
Ask it the way you'd ask your team: which tool carries the usage, how much spend rose this month or how many people use AI every week. Pulse AI answers with the chart and the table, using only the data your role can see.
Northstar Labs
Illustrative samplePulse AI
Where does AI spending go?
The example attributes US$8,120 to recorded AI consumption in the last 30 days. Most of it sits in two models.
Attributed AI cost
US$8,120
recorded consumption, last 30 days
+6%
vs previous month
Attributed cost by model
| Model | Cost |
|---|---|
| claude-sonnet-4.5 | US$4,380.50 |
| gpt-5-codex | US$2,310.20 |
| gemini-2.5-pro | US$960.40 |
| Other | US$468.90 |
Choose a question above to explore a prepared answer.
Prepared answers with illustrative data from the fictional Northstar Labs team. Choose a question to explore the example; this page does not run live AI queries.
Your existing access applies
Answers use the metrics available within your organization and team scope.
Review the sourcesInspect the numbers behind the answer
Check the period and source coverage before using a trend in your investment review.
See what Analytics measures
Answer finance without turning AI into a leaderboard of people.
Each tool's dashboard shows its own usage. Engineering metrics platforms show delivery. HubBound joins them per ticket.
| Capability | HubBound Analytics | Each AI tool's own dashboard | Engineering metrics platforms |
|---|---|---|---|
| Usage and cost by tool and model | Yesseveral tools at once | Yesonly its own tool | Partial |
| Link from session to ticket to PR | Yes | No | Partial |
| Production signals (Datadog) | Yes | No | PartialDORA, CI |
| Team experience (Pulse Survey) | Yeswith response minimums | No | Partial |
| Analysis of each AI session (summary, prompt quality, context hygiene) | Yes | No | No |
| Assistant that answers with charts based on permissions | YesPulse AI | No | Partial |
We compare tool categories; each product has its own capabilities.
Team-level by default.
Per-person views are for administrators and team leads only, and aren't meant for performance reviews. Everyone else sees team trends.
Surveys people are willing to answer.
Pulse Survey publishes a group's results only once it reaches a minimum number of responses.
Compare it with the tools you use today.
Request a demoWhat would HubBound show about AI on your team?
Answer six short questions and leave your name and work email. You get a first measurement plan right away, and the demo is built around it.
- One question at a timeChoose with a click or with the number keys, and go back whenever you want.
- A plan built from your answersYou see what to measure first and what to connect as soon as you send the request.
- Nothing is sent until the last stepYour answers stay in this browser tab until you send the request.
What CTOs ask before connecting anything.
Short answers on what Analytics measures, what it costs and what it needs.
What is HubBound Analytics?
HubBound Analytics is the part of HubBound that measures how your engineering team uses AI and what it does to delivery: AI cost per ticket, unused seats, lead time and production failures. It joins usage and cost from Claude Code, Codex, Cursor, GitHub Copilot and Antigravity with Jira, Linear, GitHub, Datadog and Pulse Survey, the team survey.
How much does it cost?
Analytics is included in the Core plan: COP 90,000 per seat per month (VAT not included), with a 5-seat minimum and 6 months of history (extendable to 24 months with Long-term metrics). Pulse AI questions are billed on demand past a usage threshold. For details, see Pricing.
Does it help decide which licenses to renew?
Yes. With the seat source connected you see paid seats with no activity and, per tool, how many sessions end in merged code.
Which AI tools does it support?
Claude Code, Codex, Cursor, GitHub Copilot and Antigravity. The data available depends on each provider, its version and how capture is configured.
How long until the first useful reading?
As soon as you connect your sources you start seeing adoption and cost. When you connect Jira or Linear, an administrator can bring in the tickets from the last six months, so delivery does not start from zero.
Does it prove the ROI of AI?
It gives you what you need to make the case properly: cost per ticket, lead time, failures and team experience. Causation takes comparing teams or periods, and Analytics shows you where to do it.
Can we compare teams or people?
Teams, yes: the team map compares each team with its peers. Per-person views are reserved for organization administrators and team leads, and should not be used for performance reviews.
Do I need every integration to start?
No. AI sessions alone show adoption and cost. GitHub, Jira or Linear add delivery; Datadog adds production stability; Pulse Survey adds team experience.
What data does HubBound collect from AI sessions?
HubBound collects AI-session metadata (tool, model, tokens, duration, tool usage) and a few trimmed samples of your prompts, never full transcripts. If your org admin turns on session analysis (off by default), those samples are processed with AI to summarize sessions, classify work and assess prompts. Each person sees their own sessions; org admins and their team leads do too.
Go deeper
Two guides for walking into that meeting with your own judgment.
Metrics
DORA metrics 2026: how to calculate all five
The five DORA metrics with formulas, an 18-deployment worked example, counting rules for staging, retries and mobile, and what AI changes.
By Angel Lindarte
11 min read
Metrics
How to measure AI developer productivity (2026)
Measure AI coding productivity with net effort per accepted change, review time, AI cost and production signals. Studies reconciled, example worked.
By Angel Lindarte
11 min read
Also in HubBound:
Walk into your 2027 budget meeting with cost per ticket in hand.
We walk you through Analytics with an example team your size and the tools you already use. You leave knowing which three numbers to bring: cost per ticket, unused seats and time to production.
