Planning in Production (PIP)

One emerging sign of AI-native orgs is that the obvious “managerial” layer is much thinner because you can leapfrog planning to execution so easily. Version 1.0 is the plan. This feels like a paradox because equally, AI work feels managerial and in fact is managerial if you look closely. There’s at once no planning and only planning.

But if anyone is “planning” an “AI project” in recognizably managerial ways, they should probably be fired. It’s AI-native when doing the MVP of the thing is easier than planning to do it. It’s not “lean” though because the early iterations that replace planning activities feel different. They feel “fat” and “waterfally” in fact. Like you’re gleefully wasting what would have been expensive resources in the past and ought to use cheaper resources like paper or classic lean hacks like dummy websites.

You no longer have to choose between Brooks (“plan to throw one away” in mythical man-month) and Spolsky (“throw one away” is a thing you should never do* because V 1.0 is full of hard-earned experiential knowledge). You can now throw away a dozen versions and never lose any earned experiential knowledge because claude will remember. If designed right, the context files accumulate metis across early throwaway versions at 100x the speed at which humans grow from young greenhorns to grizzled veterans. You start a project with claude acting like a novice, and by version 4 or 5, it’s your trusty veteran who’s been on the project way longer than you and understands it better than you, with you being the green young manager. Remember, AIs live in superhistorical time and info-age far faster than us. There is a new “thing you should never do” though, which is throw away any data/log history or archived source code. AIs can remember perfectly only if they can rewind, and raw data is always retained.

Yes, using agents is like managing knowledge workers, but you don’t manage them through plans, schedules, budgets, specs, and motivation/mentorship. You manage them in a “planning in production” (PIP) mode. That whole testing-in-production idea has now migrated way upstream to even paper-napkin and whiteboard design stages. This means any senior managerial experiences you have logged give you a serious advantage over inexperienced people, but only if you are willing to do PIP. Otherwise you’re doing the equivalent of trying to drive a car with a buggy whip. And because your “employees” work so fast, you don’t monitor and review through regular review processes. You rig the equivalent of market regulation protocols (ht Rohit Krishnan for inspiring this thought) into the scaffolding — circuit breakers, tripwires, flash-crash detectors etc. This isn’t as novel as it might seem. We already have slower versions in traditional industry. The shift from GM style don’t-stop-the-line factory supervision to Toyota lean style and on cords is a primitive example of similar thinking. A “marketizing” protocol. Any employee can stop the line by pulling the cord, and you solve for minimizing MTTR (mean time to recovery) instead of MTBF (mean time between failures). Agentic loops take this dynamic to the limit. In a sense, you’re doing only high-speed recovery, and the agents are constantly pulling the andon cord. Some are real failures, others are designed failures created by the regulatory protocols you’ve put in place. You’re not used to the tempo, but the AI does a LOT between “failures” where intervene. Between two “approve/deny” prompts for you, the equivalent of months of work happen. So your managerial tempo is not actually that weird.

You still need the paper napkins and whiteboards but you use them differently now. You use them to work out tricky framing puzzles and exploring orientation high concepts that block vibe-driven experimentation. Other whiteboard grade thinking tasks like fermi estimation, logical flows etc can likely be worked out through planning-in-production. In fact one of the toughest learning curves I’m on right now is an AFK one — unlearning bad whiteboard and paper-napkin habits and replacing them with new ones.

*