Kollabe
Retrospectives
Planning Poker
Standups
IceBreakers
Pricing
Dashboard

Posts

The Code Review Bottleneck Is Growing: What 11,000 Retro Complaints Show

Flat illustration of a long queue of paper documents waiting in front of a single small gate, with one character stamping documents while others pile up behind
Kelly Lewandowski

Kelly Lewandowski

Last updated 19/07/20267 min read

When we analyzed a million retro cards, code review wasn't the most common complaint. Testing took that crown. But code review was the complaint teams voted for most, and it had a detail we couldn't stop thinking about: over a thousand cards in the dataset are near-paraphrases of one sentence, "pull requests sitting too long waiting for review." So we went deeper. Same method as before: aggregate analysis of retrospectives run on Kollabe, columns classified positive or negative by name, themes matched by conservative English keywords, no individual card quoted or identifiable. That surfaces 11,246 code-review complaint cards written between October 2023 and July 2026, and one finding that should worry engineering leaders in the AI era.
11,246

code review complaint cards

+30%

growth in complaint share since 2024

0.75

votes per card — highest of any theme

The only pipeline complaint that's growing

Most complaint themes hold a steady share of retro boards year over year. Testing, the volume leader, has sat between 10% and 12% of complaints every quarter since early 2024. Code review is the exception:

2024 average

3.8%

2025 average

4.1%

2026 so far

4.8%

Share of all complaint cards mentioning code review, PRs, or reviewers: up roughly 30% from the 2024 average, climbing steadily quarter over quarter, and touching 5% in mid-2026. One in eleven retrospectives now contains a review complaint. We can't prove the cause from card counts. But the timing matches the one thing that changed most about how software gets written: AI coding assistants raised the volume of code arriving at the review gate, while the number of humans on the other side stayed the same. The complaint that grows is the queue. We wrote about the wider pattern in why retrospectives matter more with AI coding tools.

What teams actually say about review

Within the 11,246 cards, the sub-themes rank like this (buckets overlap; keyword floors):

Waiting — PRs sitting too long

10.4%

Feedback quality and nitpicks

6.3%

Reviewer availability

4.8%

PRs too big to review

4.2%

Review prioritization

2.5%

Merge conflicts from stale PRs

1.0%

Waiting dominates, as expected. The surprise is in the votes:
  • PR size: 0.99 votes per card. The most-voted review sub-theme, half again above the 0.65 baseline. Teams know oversized PRs are the upstream cause of everything else on this list.
  • Feedback quality: 0.77. Nitpick wars and drive-by comments draw agreement fast.
  • Waiting: 0.73. Written most, voted slightly less — everyone already knows.
There's a hopeful signal in the positive columns, too: 7,943 praise cards mention review. For every three review complaints, two cards thank a reviewer or celebrate a smooth review cycle. Teams notice when review works.

Why review complaints deserve extra weight

Votes are the retro's honest signal. Anyone can write a card; votes cost scarce attention. A theme that is both rising in volume and leading in votes is the definition of a growing bottleneck, and it's the kind of thing that hides in plain sight because review delay feels like weather rather than a decision.

Breaking the bottleneck

Flat illustration of a large paper document being cut into several small tidy documents that flow quickly through multiple open gates The sub-theme votes point at the fix order:
  1. Shrink the PRs first. It's the highest-voted complaint for a reason: small PRs get reviewed faster and better, and they go stale less often. If stories keep producing thousand-line PRs, the problem starts in refinement — splitting stories properly is review hygiene in disguise.
  2. Make review latency visible. Teams argue about anecdotes until someone measures time-to-first-review. Flow metrics turn "PRs sit forever" into a number that can go down.
  3. Agree on a first-response window. Not a full review, a first response. Most of the pain in the waiting cards is the silence, not the total duration.
  4. Close the loop in the retro. A review complaint that appears two retros in a row has stopped being feedback. Give it an owner and a due date — owned action items complete 2.7x more often.
If review keeps surfacing on your board, run the next retro with a column dedicated to it and vote honestly. The retrospective template generator can build a review-focused format in a few seconds, and the full aggregate numbers behind this post live on our retrospective statistics page.

Yes. In our data from 87,719 retrospectives, code review's share of all complaint cards rose from a 3.8% average in 2024 to 4.8% in 2026, roughly 30% growth, while other pipeline themes like testing stayed flat.

Waiting. About one in ten review complaints explicitly mentions PRs sitting too long or stuck. But the most-voted complaint is PR size, which suggests teams see oversized PRs as the root cause.

The timing is suggestive: review complaints climbed steadily through 2024-2026, exactly as AI assistants increased the volume of code reaching human reviewers. Card counts can't prove causation, but review is the only delivery-pipeline theme growing in our data.

Aggregate, anonymized analysis of 1,081,076 cards from 87,719 sprint retrospectives run on Kollabe between October 2023 and July 2026. Columns classified by name, themes by conservative keyword matching. No individual card is quoted or identifiable. Queried July 19, 2026; the original analysis used a July 14 snapshot, which is why its totals are slightly smaller.