About
Your team is humans and AI agents. Your tools only see half of them.
WhoWorked tracks human hours and AI agent contributions on one timesheet, so you can bill accurately, plan capacity honestly, and prove what your team actually delivered.
Why this, why now
Two categories, both incomplete.
Time tracking was built for a world where only people did the work. Every tracker on the market still counts that half well and has nothing to say about the other one. Meanwhile the tools that do watch agents speak in tokens, traces and spans, which is a vocabulary nobody has ever put on an invoice or taken into a budget review.
So a firm delivering with both ends up reconciling two systems by hand, or more often not reconciling them at all. Neither category is wrong. They are each missing the half the other one has, and the join between them is where the questions that matter now get asked: what did this cost to deliver, what should it be worth, and who actually did it.
We are not here to replace the tracker your team already knows how to use. We are here because the record it keeps stopped being the whole record.
What we believe
01
Attribution is credit, not surveillance.
Delegating to an agent is a manager delegating to a junior. The person who directed the work gets the headline, and the agent session is the detail underneath. Every downstream decision in the product follows from that one, which is why our reports lead with what someone delivered rather than what a tool observed about them.
02
A claim you cannot check is not a claim.
Most AI policies read well and prove nothing, because no clause produces a record. We build the other way around: each thing you assert to a client should be paired with the entry, session or export that stands behind it. It is a slower promise to make and the only one worth anything in a rate conversation.
03
An instrument has to be able to disappoint you.
A tool that can only report good news is a marketing asset, not an instrument. So our leverage calculator returns numbers below 1.0 when the inputs call for it, and that is not a hypothetical setting: a randomised trial found experienced developers took 19% longer with AI while believing they had been 20% faster. If AI is costing you capacity, we would rather you found out here.
Who it is for
The same blind spot, three different bills to pay.
Agencies and consultancies
Your clients have started asking how much of this was done by AI, and the honest answer is worth more than the flattering one. Written, as it happens, by people who spent years billing clients for time.
For dev studiosEngineering leaders
Agents do real work on your codebase and none of it reaches your tracking. We translate sessions, tokens and completions into the units the rest of the business argues in: hours, contribution share, cost per deliverable.
For IT services and consultanciesOps and finance
You are staffing and budgeting against half the data. One view of total output, human and agent, is the difference between a capacity plan and a guess that has been rounded confidently.
See what it tracksThe team
Built by a small team of humans and AI.
Between us: years of co-founding and running a client services business, where the work was sold by the hour and the utilisation question was a monthly argument long before AI complicated it. First product hire at a company that grew from a handful of people to several hundred. First engineer on a platform that went from nothing to serving institutional clients inside a year.
Today we both spend our working hours on the same problem from the inside, deciding where AI genuinely changes how a company delivers and then having to prove it afterwards. We built WhoWorked because we needed that proof ourselves and could not find a tool that would give it to us.
- Contribution mix
- With an agent
- 57%
- Human only
- 43%
- Leverage ratio
- 2.8x
AI attribution for this site
AIA HAb Ce Nc Hin R Claude Opus 5 v1.0
This work was a blend of human and AI contributions. AI was used to make content edits, such as changes to scope, information, and ideas and make new content, such as text, images, analysis, and ideas. AI was prompted for its contributions, or AI assistance was enabled. AI-generated content was reviewed and approved. The following model(s) or application(s) were used: Claude Opus 5.
Build your own statementFind out what your own record would survive.
Start tracking both halves of your team, or take the shorter route: build the AI disclosure you would be willing to hand a client, and see which clauses you could evidence today.