The Team Spends the AI R&D Budget. Why Does One Person Get Stuck with the Write-Up?

More teams issue AI plans at the team level now. Some roll out $20 plans to everyone; others distribute max plans to a few for experiments. When I got mine, I was excited. Compared models, tested harness configs, ran parallel work.

The problem came after. Received the max plan, so sharing what I learned felt obvious. I shared. Response wasn't what I expected.

In the previous post I wrote that consuming a max plan and operating one are different. This post extends that: when synthesis work piles onto the operator, it's not a personal problem — it's a structural gap the team hasn't designed for.

Distribution exists, return doesn't

The organization distributes plans. But "who synthesizes, who shares, who fields follow-up questions" is rarely designed.

Synthesizing experiment results often takes longer than the experiment itself. Anyone who's done it knows. A 30-minute experiment can take two to three hours to write up so others understand. Add slide decks, context explanations, and fielding questions — longer still.

When ownership of this process isn't assigned, the experimenter absorbs it all. They experiment, write up, present, answer questions, handle follow-up requests. Nobody asked them to. It just happens.

Sharing burns out for structural reasons, not laziness

At first you share gladly. It's interesting, you want to spread the word, returning knowledge feels right after receiving a max plan.

Over time the dynamic shifts. Sharing draws silence, not feedback. Hard to tell if anyone read it. Then the max plan itself becomes a redistribution target. "Let's all use team plans fairly." The experimenting, synthesizing, and sharing — none of that counts. Just "funny how you volunteer for the annoying stuff."

Subtler things happen too. Shared write-ups get treated as "that person handles it." Skip a round and "why'd you stop?" appears. Voluntary work quietly became mandatory.

Share and the work grows. Don't share and nobody notices. When this repeats, rational people share less. Not a willpower failure — a predictable outcome of zero incentive.

What designing return looks like

A return structure doesn't require a big system. A few deliberate choices change the dynamic.

Breaking the habit of "the experimenter always presents" already reduces burden. Rotate presentation and documentation ownership — let someone else synthesize while the experimenter hands off raw results.

Separating experimentation from synthesis matters too. A experiments, B synthesizes, C presents. The experimenter focuses on experimenting. The synthesizer learns by writing it up.

Ultimately, "who received the benefit" and "who does the return work" need separate tracking. When people who consume experiment results don't contribute to synthesis, they receive benefits without bearing costs. Just being aware of this rebalances the team.

For operators to keep operating

Individual excellence isn't enough. Return needs structural design to sustain.

What I want isn't grand. Someone experiments and writes it up. The team reads, gives feedback. Next time someone else takes the write-up. Everyone pitching in together. Without that, experimenters burn out, sharing drops, and the team loses its learning loop.

R&D support doesn't end at distributing budgets. The loop where the team absorbs experimental knowledge together — that's what makes support work.