Cognitive Surrender: Addy Osmani on Thinking When Agents Work

The warning in the essay

Accepting AI output without review is a form of cognitive surrender.

Addy Osmani’s long essay, discussed in English as “Critical thinking when agents do the work,” is a caution about a habit, not a new tool launch. The core claim is simple: when an agent produces the draft, the plan, or the code, skipping your own review and taking the result as-is is cognitive surrender.

So far he has only shown the piece in a short preview and announcement. No public build, download, or official page has been published. What is circulating is the argument itself: agents can do the labor, and you still have to do the thinking.

Why ordinary users and developers care

Agents now sit in the middle of ordinary work. You describe a task, they write the first version, and the tempting next step is to ship it. Osmani’s point lands because that shortcut looks like productivity. The cost shows up later, when a wrong assumption, a quiet bug, or a polished but hollow answer passes because nobody slowed down to judge it.

For developers, the essay is a reminder that delegation is not the same as sign-off. The agent can search, draft, and revise. You still own whether the result is true, safe, and actually what you asked for. For everyone else using assistants the same way, the same split applies: let the agent work, then read what came back as if a fast colleague handed you a draft.

• Review is the job that remains

  The essay treats unexamined acceptance as the failure mode, even when the output looks finished.

• Speed is the trap

  Agent drafts arrive quickly, which makes skipping the read feel efficient until a bad result ships.

• No product to install

  This is an argument from Addy Osmani, not a downloadable checker or a new agent you can run.

How people are reading it

The piece is drawing two familiar reactions. One side hears a needed brake: if agents write more of the work, human review has to get stricter, not looser. The other side worries the warning can be heard as a call to distrust every draft, which would erase the reason people use agents in the first place.

Neither reading is settled by a product demo, because none exists yet. The useful middle is already in the title’s idea. Agents do the work. Critical thinking stays with the person who will be accountable for it.

What is still open

The announcement does not spell out a checklist, a tool, or a rollout. Open questions sit right on the surface. How much review is enough when an agent touched every file? Who is responsible when a team accepts a draft nobody fully read? And does “critical thinking” mean a full reread, or a sharper set of checks aimed at the risky parts?

What to watch next is whether Osmani expands the essay beyond the short preview into a full public page, and whether teams treat cognitive surrender as a named risk in how they use agents. Until then, the practical takeaway is already available: do not take the agent’s output as finished just because it arrived complete.