The Biweekly Review: Persevere, Pivot, or Kill

Every two weeks, each project answers one question: persevere, pivot, or kill. Killing is the expected output of rapid prototyping, not personal failure.

How fast you ship depends half on the model and half on how fast you dare to kill what is not worth building. Anthropic Labs' practice krieger-anthropic-labs: every project goes through a review every two weeks with exactly three options — persevere, pivot, or kill. Rapid prototype, internal delivery, early access, wind down if it fails: four stations on one pipeline, not four destinies.

After reading, you should be able to answer:

• Why does a biweekly review build psychological safety instead of destroying it?
• Why should the org chart not be welded to projects?
• When should you unship a feature that already launched?

Killing is an expected outcome, not a failure

The point of the biweekly review is not "reviewing accurately"; it is making the kill the process's expected output. A project getting killed means rapid prototyping is working — it saved you three months of sunk investment. After a few rounds the team develops psychological safety: nobody treats their own project as their life, because everyone knows most bets are supposed to die krieger-anthropic-labs.

This is the other face of iterate fast, do not wait for perfect: fast iteration requires cheap rounds, cheap rounds require acceptable failure, and acceptable failure requires that getting killed carries no blame. Without the last link the rest is empty talk — the team will drag every project into a zombie with "give it two more weeks."

Bet teams: people follow bets, not projects

Once the review cadence speeds up, the org chart cannot reorganize around projects — you would be reorganizing every two weeks. Anthropic's answer is bet teams krieger-anthropic-labs: assembled across disciplines, with a bet lead as the directly responsible person who does not manage everyone on the team; only after a product proves durable does it harden into a dedicated team. Engineering management's value in the org does not disappear — coaching, collaboration, and personal growth remain core; what changes is the old equation "team = project."

It applies to a small squad too: when you are running several ideas at once, do not staff each one for the long term; concentrate attention on whichever idea enters validation, and pull it back if validation fails. People follow bets, and attention stays reallocatable.

Dare to unship the last generation's primitives

The biweekly review governs "keep building or not"; a second discipline governs "keep it or not": unship — dare to delete what has already shipped krieger-anthropic-labs. Anthropic keeps an internal channel dedicated to discussing what to delete, and they have really deleted: the Styles feature was cut for low usage and being too prescriptive — Skills is the better shape. The previous generation of AI product primitives becomes this generation's complexity debt.

The bigger problem than any single feature is complexity across surfaces: users cannot tell where one product form ends and the next begins, and flows that do not interoperate and force manual copy-paste should not exist. Adding feels productive, but a product's lifespan depends on your willingness to subtract — at review, ask "has this feature been live two months, and is it still used?" along with everything else.

Psychological sustainability is the management question of the AI era

The pace itself is a new management variable krieger-anthropic-labs: between the weekly all-hands and Wednesday, a "weekend AI" can appear — a competitor ships a model, a product, a regulation change; the rhythm runs far faster than before. Two habits hold up. The sports lens: a team is never as good as its best day or as bad as its worst, mood cycles recur, and remembering that gives perspective. And saying the mood out loud: what one person is feeling, others on the team are often feeling too — saying "I am sad about this project" is what opens the room to discuss what comes next.

This sits last not because it is soft. Biweekly reviews, daring to kill, daring to unship — each draws down the psychological account; sustainability is not a perk, it is the precondition for this cadence to keep running.