Delegation Judgment¶
The Bottleneck Is You¶
Look at how you've been working: write a user story, hand it to AI, wait for the output, verify, move on. That pipeline works. You proved it in Lessons 1 through 3. But it's serial. One thing at a time. And AI builds fast enough that you spend more time waiting for yourself to move on than waiting for AI to finish.
The temptation is to throw everything at AI in parallel and see what comes back. Don't. Parallelization amplifies whatever you put in. If you delegate well-specced work in parallel, you get multiple verified features back. If you delegate fuzzy work in parallel, you get multiple outputs that don't integrate, don't match your intent, and take longer to fix than they would have taken to build sequentially.
Scaling requires judgment first, speed second.
The Delegation-Ready Test¶
In Lesson 1, you learned: "If you can't write acceptance criteria for it, you're not ready to delegate it." That principle is even more important when running things in parallel, because you won't be watching every step.
Here's the test:
| Question | If Yes | If No |
|---|---|---|
| Can you write 2-4 specific acceptance criteria? | Delegation-ready | Needs more Explore/Plan work first |
| Is the scope bounded - you know where it starts and stops? | Delegation-ready | Decompose further (Lesson 2) |
| Can it be built and tested independently of other in-progress work? | Safe to parallelize | Needs to be sequenced, not parallelized |
| Would you know a good result if you saw one? | Delegate with confidence | Spend more time understanding the problem |
This isn't new. It's the same judgment you've been building all day. Right-sizing from Lesson 2. Acceptance criteria from Lesson 1. Independently shippable pieces from Lesson 2's decomposition. The difference is that when you're running three things simultaneously, a fuzzy spec doesn't just produce one bad output. It produces a bad output that conflicts with the other two.
What's Ready?
Team Discussion | ~4 minutes total
Your Final Sprint challenge includes outcomes like: the analyst can tell whether an AIS gap has a legitimate safety explanation by checking the piracy threat level for that region, the analyst can compare a vessel's stated destination against its actual heading to spot deceptive behavior, all signals (sanctions, gaps, satellite correlation, flag state, piracy context) combine into a single risk score the analyst can triage from, and an incoming analyst or partner agency can read an AI-synthesized summary and immediately understand the risk picture.
These are outcomes, not features. Most of them are not delegation-ready yet. Discuss: Pick one and talk through what you would need to decompose it into before each piece passes the delegation-ready test. Which of those pieces could run in parallel, and which need to be sequenced?
Key Insight
Scaling delegation doesn't mean delegating everything at once. It means knowing what's ready. The same judgment you've been building since Lesson 1 (can I spec it? is it bounded? would I know a good result?) becomes your filter for what to run in parallel. If you can't spec it, AI can't deliver it, and that's doubly true when you're not watching every step.