Codegen, Inc. - a ClickUp Company

The right search strategy depends on whether the user knows the vocabulary of their own data. Jeff Huber from Chroma breaks down why lexical search (like regex) works so well for code retrieval: LLMs have been trained on enough code that they're good at guessing the exact terms to search for. In the latest AI Hot Takes, Jeff explains: ☑️ If you're a subject matter expert in your data, lexical search is fast and precise ☑️ If you don't know the vocabulary, semantic search helps bridge the gap ☑️ The blend varies by application — some are 80/20 lexical, others flip to 20/80 semantic ☑️ Code agents succeed with regex because models understand code grammar exceptionally well Check out the full conversation here: https://lnkd.in/d6-44fWR

When customer requests pile up faster than engineering can respond, something has to change. Warmly customer success managers used Codegen to turn insights into production-ready features — without adding to the dev backlog. maximus greenwald, co-founder and CEO, said: “A CSM basically can act as a product manager and then pair with an engineer and actually get stuff done. We have a ton of PRs already out to staging and we're going to be deploying.” For Warmly, the result was faster delivery, tighter feedback loops, and a whole new way to collaborate across teams. See how they did it 👉 https://cdgn.sh/36OGCa

This week’s updates focus on polish and performance — making Codegen smoother, faster, and more intuitive for developers. ☑️ Unified scrollbar system across 20+ components for a consistent UI. ☑️ Kanban upgrades with tri-state column controls ☑️ Compact date formatting for cleaner analytics. ☑️ No more GitHub upload limits — large files now supported. ☑️ Jira issue linking for programmatic relationship management. ☑️ Slack audio message support for voice-based agent interactions. Small refinements. Big gains 🦾 Explore what’s else is new → https://cdgn.sh/oStmbT

AI development is moving past one-off coding bots — the real leverage now comes from orchestrating many agents at once with parallelism, clean states, and human oversight. ☑️ Specialized agents got us off the ground ☑️ General-purpose models like Claude Code & Gemini CLI took over ⚡️ Now the future is orchestration As Louis Knight-Webb said on AI Hot Takes: “The human element of review is very difficult to replace…I don’t think it is going to go away anytime soon.” That’s why Codegen is building the orchestration layer for the next era of AI development. 📖 Read the full deep-dive here: https://cdgn.sh/eUi1WL

Lambda Curry uses Codegen to extend their team inside GitHub, Slack, and Linear. Instead of bouncing between tools, their developers can: ☑️ Generate and refine new features directly in GitHub ☑️ Make small changes instantly without slowing down ☑️ Ask Codegen to explain complex code and dependencies ☑️ Open and track Linear tasks without leaving the workflow As Lambda Curry Co-Founder Jake Ruesink put it: “It helps us go from task planning to implementation with astonishing speed. Sometimes what we thought would take hours now takes minutes. It’s become hard to estimate timelines — in the best way.” Read the full story: https://cdgn.sh/UzST55

Last week we shipped upgrades that make Codegen faster, more resilient, and ready for broader enterprise use: ☑️ New Kanban Dashboard – feature-flagged rollout, smoother drag-and-drop, tighter type safety ☑️ Repository page fixes – resolved hydration issues for rock-solid loading ☑️ PR review improvements – cleaner defaults, fewer surprises ☑️ Sandbox expansion – multi-provider architecture (Modal, CodeSandbox, Runloop, Codegen) for flexible dev environments ☑️ GitHub integration – agents can now upload artifacts up to 100 MB directly to a repo Plus sharper analytics and refined pagination for a smoother UI. Codegen is building the OS for code agents — and every layer just got stronger. 👉 Read the full weekly diff here: https://shorturl.at/eFJUG

Coding agents hit $600M+ ARR while other AI applications struggle to find ground. Jeff Huber from Chroma explains that coding became the dominant AI use case because software development provides natural constraints that make LLM unpredictability manageable. In the latest AI Hot Takes, Jeff explains the current limitations holding back even bigger wins: ☑️ SWE-bench uses relatively small codebases compared to production reality ☑️ Real systems like 90 million token TypeScript monoliths present unsolved challenges ☑️ Code retrieval needs to evolve beyond current capabilities to handle enterprise scale ☑️ The deterministic software environment makes coding the ideal testing ground for AI The success of Cursor and Claude Code wasn't accidental — coding provides natural constraints that make LLM unpredictability manageable while delivering measurable value. Check out the full conversation here: https://lnkd.in/d6-44fWR

🚫 Most productivity stacks promise speed and deliver… another tab. The teams that actually move faster do three things well: keep the AI in the loop where work happens), treat CI like a product, and track statistically grounded metrics. If you want a concrete path, pair an assistant with a review agent, add a check-suite fixer, trigger tasks from Slack/GitHub, and feed real metrics back into planning. Read the full deep dive and discover how Codegen helps 🦾 https://shorturl.at/m5DJa

Chroma published research showing models degrade as early as 5,000 tokens and the labs aren’t being transparent about this. Jeff Huber breaks down context rot — the gap between what labs advertise and what actually works. Currently developers are reporting things getting squirrely past 100K. In the latest AI Hot Takes, Jeff explains why this isn't going away: ☑️ Degradation appears fundamental to transformers and model training ☑️ Steep drops in attention and reasoning happen much earlier than advertised ☑️ Context engineering will be essential for the next 3-5 years minimum ☑️ Speed and cost matter as much as accuracy in the "make it work" phase Chroma plans to run the same analysis on GPT-5 and the new million-token Claude model. Engineers need to work around these limitations rather than rely on raw context window size. Full episode: https://lnkd.in/d6-44fWR

🚀 Codegen On-Prem Software Is Here Enterprises can now deploy the full Codegen platform inside their own infrastructure — with all the scale and reliability of our cloud offering. What’s inside: ☑️ Kubernetes-native architecture with ready-to-use Helm charts ☑️ Complete data sovereignty and custom API key management (model config docs) ☑️ Flexible deployment from standard clusters to air-gapped environments ☑️ Security & compliance built for enterprise audits ☑️ AWS Marketplace listing coming soon Get the same developer-first experience with the control, isolation, and enterprise support your org requires. Explore everything we have to offer 👉 https://shorturl.at/6Lc6x