The Observatory

If you're shipping agents before you've stress-tested what they're actually supposed to do, you're burning compute on the wrong problem. Maurya's method tells you exactly how to find that out faster.

If your agents are shipping work that feels a little ragged and undersized right now, that's not a bug in your process — it's the process. Founders at Work should be required reading before you hire your first human.

If you're raising money for an AI-native company, the leverage you think you have from moving fast with agents still gets adjudicated on a cap table. The game has the same rules it always did.

If you're handing work to agents before you've done that work yourself, manually and awkwardly, you don't yet know what the agents should optimize for. That's the problem Paul Graham's essay quietly solves.

If your agents are doing the work, the human still has to choose to come back — and that return visit is a design problem, not a marketing problem.

Ben Horowitz wrote his book for founders who discover, usually too late, that no framework survives contact with a genuine crisis — and that discovery hits differently when half your workforce is software.

If your agents are just chasing one-off outcomes, you're doing goal-setting, not system-building — and James Clear spent a whole book explaining why that distinction kills you slowly.

If you're deploying agents into a marketplace, a collaboration tool, or any product where value compounds with participation, you have a cold start problem — and wishing for traction is not a strategy.

Cagan's core warning lands harder when agents can execute faster than any team can think — output velocity is not the same as product value, and now you have no excuse to confuse the two.

If your agents are shipping output no customer ever validates, you're not moving fast — you're just accumulating waste faster than before. The core discipline here is more urgent than ever, not less.

Kahneman's core finding — that fast, confident judgment is the default and slow deliberate review is the exception — maps almost perfectly onto how language models behave in production. If you haven't designed the audit, it isn't happening.

If you are building with agents, you already solved the supply side — work happens automatically. The demand side will still kill you, and Weinberg and Mares wrote the map.