Seven-week cycles. Five weeks heads-down on roadmap work — shipping features, fixing bugs, the object-level stuff that moves the business. Then a two-week exploration block where the team deliberately stops executing and instead asks: what’s changed in the world, and does our stack/workflow/toolkit still make sense?

The two weeks aren’t vacation and aren’t “20% time.” They’re structured differently from execution weeks but they’re still full-intensity work. Roughly:

Week 1 — Scan and prototype. Everyone picks 1-3 things to seriously investigate. New models, new libraries, new dev tools, new workflow patterns, a competitor’s approach, a paper, an internal pain point. The deliverable isn’t a PR — it’s a working prototype plus a writeup. “I rebuilt our eval harness on top of X. Here’s what broke, here’s what got 3x faster, here’s what I’d need to migrate for real.”

Week 2 — Converge and decide. Team shares findings. Some things get adopted wholesale (“we’re switching our matcher to this”). Some get queued as backlog items for the next execution block. Some get killed with a short postmortem so nobody revisits them in three months. The block ends with a written decision log and an updated “stack of record.”

Crucially: no roadmap commitments are allowed to bleed into the exploration weeks. If a VP says “can you just squeeze in this one thing” — no. The whole point is that the time is sacred, otherwise it collapses into execution and you’re back where you started.

20% time gives individuals permanent slack to pursue side projects. Nice in theory, but two failure modes:

1. It quietly evaporates under execution pressure because it’s per-person and nobody defends it.

2. The work is individual, so learnings don’t propagate to the team. 5/2 is collective and time-boxed — everyone stops together, everyone shares back, decisions are made at the team level. No diffusion of responsibility.