This is the direction we believe in at Lightsprint: agents should sit inside a shared shipping loop, not another chat box. Product intent, code-aware planning, parallel execution, previews and engineering review belong in one surface.
- This is the benchmark shift that matters: test agents on your own work. Public scores miss task mix, cost, latency, reviewer corrections and whether the output survives merge. Internal evals are becoming company infrastructure.At Databricks, we built an internal coding benchmark on our multi-million-line codebase. Biggest takeaways: 1. Open-source coding models are no longer toys. GLM-5.2 is strong and cheaper. 2. OpenAI, Anthropic, and GLM-5.2 are all on the Pareto frontier. The real unlock is an
- Exactly. Skills matter when they package judgment, not just prompts. Models change quickly; good ways of scoping, reviewing, debugging and verifying change much more slowly. Skills are how those practices become portable.Why I think skills are worth bothering with:
- By far one of the most common questions about @lightsprintai is how merge conflicts are handled at scale. We thought so too! But counterintuitively, the agent actually is able to handle it easily. thats because it can easily figure out with the context of both authors!
- Excited to launch this today! Talk to us if you'd like to move quickly and ship things as a team :DWe’re launching the missing team layer for AI-powered development 🚀 The SDLC was built around coding. Coding has changed, and the SDLC must be updated. Lightsprint makes agentic development collaborative with visual plan mode, parallel cloud agents and PR preview
