blog: three calls we'd have gotten wrong building our background agent - #239
Merged
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
ishaan-berri
changed the title
…inks/typos
- Move "What we shipped" above the numbered learnings so the 30% proof lands early
- Tie the ~30% claim to weekly ticket throughput; reframe 138 closed-without-merge as the cheap-session design, not noise
- Convert numbered sections to H2 headings for sidebar nav
- Add {/* truncate */} fold marker
- Fix broken author-PR link and "its PRs"/"its memory"/"smart enough" typos
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
- Lead with the hook (21 merged, no human code + the MITM story) instead of a meta intro - Tie the ~30% claim to eng tickets/week and restore the "138 closed by design" framing so the merge ratio reads as intent, not noise - Embed the two staged assets: Slack cold-start screenshot and the brain/sandbox split diagram (the diagram also documents the no-BASH design the description promised) - Add real artifacts: lite-harness repo contract and the credential-to-host vault config that kills the MITM - End each section on a claim instead of a hedge; soften the demo CTA Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
- Replace hook with user's framing: "goal to 10x productivity" → "30% of tickets" - Add required sections: Key Takeaways, FAQ (w/ OSS vs Enterprise), Conclusion, Recommended Reading - Add ## Where the AI Gateway fits (required H2 + brings AI Gateway to 5 mentions) - Update frontmatter: description w/ CTA hint, ai-gateway tag, hide_table_of_contents: true - Remove rogue-agent lede (trust risk); MITM stays as §3 lesson Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Remove sandbox_read_file and sandbox_write_file; keep only sandbox_provision and sandbox_execute. Reposition boxes centered, update arrows, label, and matching prose. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…ence) Shows Ishaan's real-time Slack messages catching the agent circumventing the credential stub vault and storing real keys in its memory. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Fix ambiguous OOM attribution in §2; fix "it's"→"its", "grew slower"→"grew more slowly"
- Tighten "unifies X to Y" → "adapts X to a single HTTP contract"
- Replace "big win in terms of" with concrete sentence
- Fix dangling -[blog] link; clean up admonition block links
- Fix §3 closing: "Our learning here was that" → "The lesson:"
- §4 (AI Gateway): remove filler opener, fix space-before-comma, fix "Furthermore"
- §5: "is the part that is mostly solved" → "is largely solved"; "do destructive actions" → fix; "MCP's" → "MCPs"; "more preferable" → "more reliable"
- Key Takeaway: fix contradiction with §3 on LLM-level guardrails
- FAQ: deduplicate Cursor answer; fix "self-hosted" as license attribute
- Try it: consolidate duplicate links from info block into one clean sentence
- Fix stale SVG alt text ("four tool calls" → "two tool calls")
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Colons for appositives, commas for conjunctions, parens for asides, periods for sentence breaks. Zero em dashes remain. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
krrish-berri-2
enabled auto-merge (squash)
May 28, 2026 04:12
krrish-berri-2
approved these changes
May 28, 2026
6 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
A blog post about what we learned shipping a background coding agent on LAP — anchored around three design calls and the Slack cold-start scar that motivated them:
Honest numbers: 21 merged PRs, 41 open, 138 closed-not-merged, 50+ filed/month on `BerriAI/litellm` over the last 3 weeks. Plus the `#help-litellm` Slack triage volume.
Includes a real Slack screenshot of the 2-minute cold-start failure mode that motivated the brain/sandbox split.
Files
Test plan