Always inspired by what devs build with Refact.ai Agent. While AI does the coding, people can focus on creativity.

At yesterday’s Weekly Showcase:

🏋️ Fitness app w/ AI motion tracking + Web3
🧠 Explainer tool w/ dev-topic animations

Watch on YouTube:
youtu.be/kTbNQsmfT08

❌Prompt engineering isn’t about magic words.
✅It’s about giving your AI Agent the right setup: targeted context, clear intent, and an iteration-friendly workflow.

Watch our webinar:
Prompt Engineering for Developers Without 🐂$h!†'

youtu.be/DP_yKoHeWI8?...

🔥 Refact.ai is now the #1 open-source AI Agent on the SWE-bench Leaderboard:
🔹 SWE-bench Verified → 70.4% solved
🔹 SWE-bench Lite → 60% solved

Soon: new score with Claude 4 Sonnet 🙂‍↕️

Full tech breakdown: refact.ai/blog/2025/op...

Our open-source SWE-bench pipeline [GH]: github.com/smallcloudai...

We're the new open-source SOTA AI Agent on SWE-bench Verified.
Score: 69.9% — 349/500 tasks solved.

Key tech behind the run:
• debug_script() sub-agent using pdb
• strategic_planning() tool powered by o3
• Automated guardrails that course-correct mid-run

🧵

🎬The recording is live! How to build AI Development Agent: youtu.be/VIup4zS10rs

Nick Frolov, Head of Product @Refact.ai, breaks down the tech behind truly autonomous programming systems — and how it all works inside our AI Agent.

Grab your coffee & catch the full webinar on YouTube!

🤔 Can Gemini 2.5 Pro dethrone Claude 3.7 Sonnet in coding?

We tested it on aider’s polyglot bench with Refact.ai: 82.2% score — 3rd place behind Claude’s 93.3% and 92.9%.

Solid — and available in Refact.ai now.

P.S. Next up: evaluation with Gemini-2.5-Pro-Preview-05-06🔥

Open-source Refact.ai achieves #1 on SWE-bench Lite with a 59.7% score!

Our approach: fully autonomous Agent, no user intervention needed. Just assign a task & let AI handle it end-to-end

• Claude 3.7 Sonnet — core model
• deep_analysis() tool with o4-mini — reasoning

refact.ai/blog/2025/so...

🧵