The Old Bottleneck Is Gone

For decades, writing code was the slowest part of software development. Teams built elaborate rituals around that constraint: backlog grooming, estimation poker, sprint planning, daily standups. Every ceremony existed to maximize the value of expensive, scarce engineering time.

That constraint has evaporated. AI-assisted coding tools have made generating code fast and cheap. The bottleneck has moved. The expensive part is now deciding what to build, giving the right context, validating results, and coordinating humans.

Yet most teams still operate as if writing code is the bottleneck. They spend hours debating story points, stack-ranking backlogs, and reciting status updates. These rituals have become dogma.

What to Do More Of

Pair Prototyping

Instead of the old waterfall of PM → designer → engineer, put all three in a room (or Zoom) with an AI coding tool. Build a rough, working prototype together in an hour. The PM sees edge cases, the designer spots interaction breaks, the engineer identifies genuinely hard parts. Then throw the code away. Use the experience to inform the real build. Code is cheap now; learning is what matters.

Connection Time

Standups were never a good way to build team cohesion. Replace them with optional, low-pressure social time. Let people connect without the Jira script. It's better for morale and doesn't waste time on status updates that nobody listens to.

Show and Tell

Experimentation is frictionless. Build things just to see if they work. Show off prototypes, even messy ones. It sparks ideas and cross-team inspiration.

Retrospectives

With AI tools evolving monthly, the trade-offs are unclear. Regular retros let teams reflect on what's working, what's not, and what opportunities exist. Double down on them.

What to Tweak

Backlog Grooming

The old goal was a perfectly prioritized backlog of tiny tickets. The new goal is shaping work so it's shovel-ready. Prioritization matters less than actionable context. Focus on making tasks clear enough to ship, not on stack-ranking every item.

Documentation

Documentation isn't just for humans anymore — it feeds AI agents' context windows. Keep it up to date. Create an automated agent job that checks for contradictions in your docs. Humans still make decisions, but agents do the grunt work.

Code Review

AI-assisted code review is a superpower. Use specialized sub-agents for security, performance, accessibility, and correctness. Frontier models catch things humans miss. Combine AI review with expert human review for better code quality than before.

What to Drop

Estimation and Sprint Planning

These ceremonies answer "how much can we build in two weeks?" But in the time it takes to debate Fibonacci numbers, you could have built the feature. Drop them. Explore alternatives like Kanban or Shape Up. Or invent something new.

Standup

The daily status update is one of the most expensive ways to solve coordination drift, visibility, and blockers. Coordination drift is better handled by pair programming and tech leads. Visibility belongs in your task tracker. Blockers should be unblocked immediately, not at the next standup. Replace standup with an open agenda meeting that only happens when there's something to discuss.

Question Everything

Periodically revisit every team ritual. Ask: What problem does this solve? Is the problem still worth solving? Is this still the best way? If the answer is no, kill it. Don't keep zombies around.

The teams that adapt fastest will be those willing to let go of old habits. The bottleneck is judgment now. Spend your time there.