AI Productivity Needs Evidence, Not Token Theatre
2026-05-26 · 4 min read · Janaina Maia
The most revealing AI workforce news this week is not that another company is cutting roles. It is how the company is explaining the cut.
TechCrunch reported that ClickUp laid off 22% of its workforce while positioning the move as a shift toward thousands of internal AI agents. The company says employees are now expected to direct agents, review their output, and create more value with fewer people. It is also talking about measuring value created and time saved, rather than simply counting how much AI people use.
I think this is the exact moment where product and design leaders need to get more precise.
Using AI is not the same as creating value.
A lot of organisations are still confusing AI adoption with AI productivity. Someone can generate more drafts, run more prompts, or automate more small tasks without improving the quality of the work. Activity can go up while judgement goes down.
This is why I do not like shallow AI metrics. Counting tokens, prompts, or tool usage is a bit like measuring collaboration by counting Slack messages. It tells you something happened, but not whether the work got better.
If AI is going to reshape teams, the evidence needs to be stronger than “people are using it.” Leaders should be asking: which decisions improved, which cycle times dropped, which rework disappeared, which customers benefited, and which risks increased?
The role shift is real, but it is not magic.
The ClickUp example points to a real change in knowledge work. In some workflows, people are moving from doing every task manually to orchestrating agents that draft, search, summarise, analyse, and prepare work for review.
That can be useful. It can also become management theatre if the organisation does not redesign the workflow around it. Directing agents is not a casual add-on to an overloaded job. It requires clearer goals, better context, stronger review habits, and explicit accountability for the final result.
The human does not disappear. The human moves to briefing, steering, checking, deciding, and owning the consequence.
Design implication: build evidence into the workflow.
AI productivity should not be measured only in dashboards after the fact. The product experience itself should make evidence visible while the work is happening.
- Show the brief: what the agent was asked to do, with what constraints.
- Show the trail: which sources, tools, and assumptions shaped the output.
- Show the review: what the human accepted, changed, rejected, or escalated.
- Show the outcome: whether the work saved time, reduced rework, improved quality, or merely produced more artefacts.
- Show the boundary: where the agent must stop because the decision needs human judgement.
Without that evidence layer, companies will reward the people who look busy with AI, not necessarily the people who use it responsibly.
My take.
The future of work will not be “employees versus agents.” That framing is too crude. The more useful question is which parts of the workflow should be automated, which parts should be augmented, and which parts must remain accountable to a named human.
AI can absolutely make teams faster. But speed is only valuable when the direction is right and the quality is visible.
If a company wants to become an AI-native organisation, it needs more than internal agents and ambitious salary bands. It needs a clear model of evidence: what improved, what got riskier, and who is responsible when the agent-assisted work becomes real.
Productivity without evidence is just noise with a better demo.