The First Rep Isn't Owning the Skill

A first freestanding handstand usually lasts a second or two and looks terrible. It still counts, but it isn’t a skill yet. A clean 10-second hold took me about a year, and for six months of that year it barely felt like I was improving at all.

That gap, between the first rep and owning the movement, is the best way I’ve found to explain the difference between merge-ready and launch-ready code.

Broad strokes come fast

With coding agents, the first version of a feature arrives remarkably quickly. Write the spec, let the agents build, gate it behind a flag, review it by blast radius (Review Depth Follows Blast Radius) and merge. That’s the broad strokes: about 80 % of the work, done in a fraction of the time it used to take.

Calisthenics has the same phase. You kick up against the wall, find the stacked line, touch balance for a second. The shape appears surprisingly fast.

Merge-ready is the first ugly rep

A wobbly two-second hold. A muscle-up with one elbow chicken-winging over the rings. It happened, and you could post a video of it, but you couldn’t do it on demand, tired, or on a bad day.

Merged code is the same. It works, the tests pass, and the PR is closed. It hasn’t yet faced real users, real load, or the rest of the product.

Owning the skill is the last 20 %

The last 20 % takes about as long as the first 80 %. In code, that’s where people’s judgement and taste matter most. Each part of it has a training equivalent:

The last 20 % in code	In training
Performance: it holds up under real load	Repeatable under fatigue: any day, not just your best rep on a fresh morning
Security: it doesn’t open a door it shouldn’t	Joint-safe form: tendons adapt more slowly than muscles, so a rep that wrecks your elbows isn’t finished, however good it looks (Deloads and Tendon Adaptation)
It meets the end user’s goals: not just the ticket, the reason for the ticket	Specificity: is the rep building what you’re actually after, or a flattering version for the camera? (SAID Principle)
Features work well together: the product holds up as a whole, not one PR at a time	Linking skills into a flow: muscle-up into dip into L-sit, or handstand into press. Each skill passing on its own isn’t the same as a routine that holds together

Why stopping at the first rep is risky

If you treat the first rep as the finished skill, you pay for it: injuries from load your tendons weren’t ready for, plateaus from grooving a bad pattern, and form that falls apart the moment you’re tired. Software shipped at merge-ready pays the same way: performance problems under real traffic, security gaps, features that technically work but don’t help anyone, and parts that clash once they meet.

The fix is the same in both. Treat the first rep as the start of the work, not the end. Practise until it’s repeatable (Handstand Technique: Base, Stack, Corrections). In code, audit the whole feature for performance, security and the user’s actual goal, then launch it gradually behind the gate so problems surface while they’re small.

Match the care to what could get hurt

The metaphor covers review depth too. You don’t warm up and check form equally for everything. A wrist-heavy press to handstand gets a careful warm-up and your full attention; a set of incline push-ups gets a glance. Code review works the same way: deep reads for infrastructure, careful reads for shared code, a proof read for isolated leaf code.

The short version

• The shape comes fast. Owning it doesn’t.
• The first rep is a milestone, not the finish line. Merged is not launched.
• The last 20 % (repeatable, safe, purposeful, connected) takes as long as the rest, and it’s the part that matters.