Building Better Agents Through End-to-End Evals
How Ramp Stack grades a bookkeeping agent: messy synthetic worlds, rubrics and skill ablation.
Curated
227 resources
AI Builder Resources
Hand-picked guides, repos, and courses for shipping with AI.
Guides
See all 140 →- Can a Local LLM Replace Claude Code? The Benchmark Everyone Quotes Is the Wrong OneWhy SWE-bench misleads, and what actually breaks when a local model drives your coding agent.
- The September Cut Took 17% of My Claude Code Week. Subagents Took 48%455 sessions of logs show where a Claude Code week actually goes, and which saving tips work.
- Evidence-Driven Development: Give Your Coding Agent Something to ProveWrite a claim that can lose, break the app on purpose, and keep evidence your agent can't fake.
- Slopsquatting: Your AI Invented a Package, and an Attacker Was WaitingNearly 1 in 5 AI-suggested packages in one study did not exist. How attackers use that.
- Free the Models: Harness Design at the FrontierReplit stopped routing work for the model and let the model decide. Here are the numbers.
- UpGuard: 16,326 Supabase Databases Exposing DataThe largest study of Supabase misconfigurations: thousands of app databases readable by anyone.
- AI Coding Made CI a Bottleneck, So Linear Reworked TheirsHow Linear cut PR wait time and halved runner cost while its test suite quadrupled.
- Jev in 25 Lines of PythonBuild a Jev-style decision model from any GGUF model in 25 lines: choices in, probabilities out.
- AI Overviews Cut CTR by 23.1% in FranceAhrefs data: the most exposed French sites lost 23.1% of their click-through rate
- We Pinned Our Model Version. The Provider Deprecated It Anyway.Why re-qualification, not inference, is the recurring cost of production AI
Github
See all 87 →Context Mode
An MCP server that keeps raw tool output out of your agent's context: 315 KB down to 5.4 KB.
- ForemergeCoding agents announce what they will change before they change it, so plans collide early.
- new-term-radarCatch new AI model, product and repo names days before keyword tools list them.
- Skill ScannerScan agent skills for prompt injection, data exfiltration and malicious code before installing.
- PulseSee how much Claude Code, Codex, Cursor and 70+ other AI coding tools you have left.
- Strands BoxSandbox your AI agent: default-deny rules for what it can run, read, write and reach.
- AOCI-CODEA Git-versioned map of your codebase that coding agents read before they touch anything.
- agentacctSee what your coding agents did and what it cost, task by task, on your own machine.
- deja-vuMemory for 38 coding agents, built from the session history already on your disk.
- lociA queryable second brain over your scattered notes and docs, with an MCP server for agents.
- BacklotLocal emulator for Slack, Gmail, GitHub, Jira and Notion APIs. No OAuth, no vendor account.