Every company I’ve talked to in the last two quarters is asking the same question about AI: what’s the ROI?
The instinct is to measure AI value in cost — jobs not hired, hours saved, efficiency ratios. That framing isn’t wrong exactly, but it misses the more interesting thing that’s actually happening. AI isn’t just making your existing workforce cheaper. It’s making your existing workforce larger. Not larger by way of headcount but how many minutes were added to the overall work output.
The Unit of Measure We’ve Been Missing
A senior engineer on your team has roughly 6 productive, focused hours in a workday. That’s not 8 hours but accounting for context switching, meetings, the cognitive cost of being a human being they will actually do about 6 “real” hours of work.
When you give that engineer an AI coding assistant that handles the scaffolding, boilerplate, and first-pass debugging, you’re not making those 6 hours more efficient. You’re adding hours. That same engineer is now operating with the output capacity of someone working 9 or 10 hours without forcing them to reduce another burner.
You didn’t have to hire anyone. You added minutes to your workforce.
Now multiply that across a team of 20 engineers. You haven’t added headcount, but your effective workforce capacity has materially expanded. The question is whether you’re tracking that expansion, and whether you’re doing anything intentional with it. The minutes get absorbed invisibly into more tickets, faster sprints, a backlog that moves slightly quicker. It isn’t a bad thing per se but it is the equivalent of getting a raise and letting the extra money disappear into your checking account.
What This Changes About How You Build
I’ve been thinking about this as the “minutes added to workforce” framing for a while now, and it’s reshaping how I think about building teams The old model was hire for capacity, then deploy capacity against problems The new model is instrument your AI deployment to understand how much capacity it’s actually generating, then make intentional decisions about where that capacity goes.
This isn’t theoretical. We have engineers who are shipping features in the time it used to take to write the spec. That delta isn’t just a productivity metric, it is a strategic resource. The question we have to keep asking is: are we allocating that resource, or just burning it?
The companies that win in the next five years won’t be the ones that added the most AI tools. They’ll be the ones that treated the capacity those tools generated as an asset to be managed and not a line item to be celebrated in a quarterly business review and then forgotten.
The Measurement Problem
Minutes added to workforce is hard to measure directly. You can proxy it with deployment frequency, PR cycle time, sprint velocity per engineer, time from idea to production. These metrics exist and most engineering orgs are already tracking them. What changes is the framing that you’re no longer just tracking speed but you’re tracking effective capacity.
If your team’s PR cycle time dropped from 3 days to 1 day after AI tooling, you didn’t just speed things up. You added roughly two engineer-days per PR to your team’s capacity.
The deeper challenge is the cognitive work that’s harder to instrument. When an AI agent handles your on-call triage and reduces the number of engineers pulled into incident response at 2 AM, you’ve added something harder to measure than PR cycle time under the umbrella of added mental energy, recovered attention and the hours that used to be lost to context-switching out of deep work.
I’ve written about this in the context of the Four Burner Problem — the idea that AI’s real value isn’t output, it’s cognitive capacity. The minutes added to workforce framing extends that: what’s true for individuals is also true for organizations.
Why This Matters More if You’re Building Something New
If you’re at a legacy company, the minutes added to workforce framing is mostly about optimization. You have existing headcount, existing capacity, and AI is expanding it. The challenge is allocation.
If you’re building something from scratch, the framing is more fundamental. You don’t start with a team and add AI tools. You start with the capacity equation and work backwards. What does your organization need to accomplish? How much of that can AI agents handle directly? What does the human layer look like around those gaps?
I wrote last week about the new math of company composition. Minutes added to workforce is part of that same equation — it’s just the supply side. You’re not starting with headcount and figuring out AI. You’re starting with the mission and designing the blended workforce around it.
The Practice
First, audit where your AI tools are actually generating capacity today: PR cycle time, incident response time, spec-to-production time with real numbers!
Second, treat that capacity as a budget. Explicit allocation beats invisible absorption. If your team generated the equivalent of two engineer-months of capacity last quarter through AI tooling, what did you do with it? What should you do with it next quarter?
Third, include cognitive capacity in the audit. Time saved on low-value work that was previously unavoidable such as ticket triage, status updates, first-pass code review, documentation. That’s not just hours recovered but it is quality of work improved and quality of life improved, which compounds in ways that are hard to model but very easy to observe.
The companies that are winning with AI right now are mostly doing this intuitively, and the ones that will still be winning in three years will be doing it systematically.
This post connects to ideas from Can AI Solve the Four Burner Problem?, Block Just Cut 3,500 Jobs. You’re Reading It Wrong., and Awe in an AI World.