Cisco DevNet's 12-engineer team ships a mandatory PubHub rewrite in Go in about five and a half months instead of nine, using Claude, Copilot and Cursor
A security mandate forced Cisco DevNet to rebuild PubHub, its internal publishing tool for API docs and Learning Labs, in Go. A hand-conversion pilot was dropped as too slow. The team then used Cursor to generate an architecture and component reference of the old app and built a proof of concept from it. Claude agents joined in June 2026, building whole features with test cases from user stories and specs, while an AI-driven framework compared old and new documentation output and Playwright automation replaced manual UI testing. The documentation lead sat in on development to keep requirements ahead of the faster engineering. The target moved from end of December to end of August, and the new PubHub reached production in mid-September 2026.
The numbers (as reported)
| Schedule | Nine planned months shipped in about five and a half (39% reduction against the pre-AI estimate) |
|---|---|
| Effort | 108 planned engineer-months of scope shipped in roughly 66, with 12 engineers |
| Throughput | 1.64x whole-project average; higher after Claude arrived in June |
| Quality | No major issues since reaching production |
Facts
- Company / subject
- Cisco (DevNet PubHub team)
- Tools
- Claude, GitHub Copilot, Cursor
- Codebase
- brownfield
- Work type
- Migration
- Stack
- Go, Playwright
- Published
- 2026-10-19
- Last checked
- 2026-10-26
What the source claims Practitioner (own blog/paper)
Claimed by the builder, writing about their own work. Not checked independently.
- Nine planned months shipped in about five and a half, a 39% schedule cut against the team's pre-AI estimate -39% schedule
“Nine months of planned work shipped in about five and a half: a 39% schedule reduction against the team’s own pre-AI estimate.”
- 12 engineers producing the output of about 20, a 1.64x throughput gain averaged over the project 1.64x
“12 engineers producing the output of about 20”
- Reached production roughly three and a half months ahead of the original December plan
Caveats
Written by the team itself. Compared against their own pre-AI estimate, not a measured baseline. Claude only joined in June, so the first two months ran at the old pace. No cost, PR or line counts. Engineering detail is promised in a separate AI for SDLC series.
More stories by tool: CursorGitHub CopilotClaude
Related topics: MigrationsBrownfield codebases