Welcome back. Every AI lab and IDE company is making the same bet: the future of engineering is managing AI agents.
The latest player in the arena is Oz by Warp: a cloud platform for running coding agents. Warp claims it already writes 60% of its internal PRs. And they’re not the only ones making moves — Prime Intellect also just made it possible for any dev team to post-train agentic models.
Today’s Insights
Powerful new upgrades and tools for devs
How to build enterprise context graphs
Fix Claude Code's memory problem
Trending social posts, coding hacks, and more
Welcome to The Code. This is a 2x weekly email that cuts through the noise to help devs, engineers, and technical leaders find high-signal news, releases, and resources in 5 minutes or less. You can sign up or share this email here.
TODAY IN PROGRAMMING
Click here to see Warp’s Oz in action
Warp debuts Oz to help dev teams run coding agents in the cloud: The AI-powered IDE maker just dropped Oz, a platform for running and orchestrating coding agents at scale. Each agent gets its own Docker environment to build, test, and write PRs autonomously. Warp says Oz is already writing 60% of its internal PRs, including a fraud-detection bot that caught nearly $60K in fraudulent usage in a single run. Click here to watch how Oz fixes bugs.
Alibaba's new image model merges generation and editing into one: The Chinese e-commerce giant shipped Qwen-Image-2.0, a single model that generates and edits images with prompts up to 1K tokens. It currently sits at second spot for image editing and #3 for text-to-image on AI Arena's Elo leaderboard. Try it here.
Prime Intellect wants to turn every dev team into an AI lab: The Silicon Valley based AI startup’s latest release introduces Lab, an environment for post-training agentic models using reinforcement learning. Instead of renting a dedicated GPU cluster, teams share infrastructure and pay per token, making it feasible to fine-tune large scale models. It supports base models from Nvidia, Meta, Qwen, and more. Start your first training run here.
PRESENTED BY GLEAN
Join Glean’s launch event to discover their latest-gen assistant: an AI work partner that actually helps all teams get things done. Hear from technology experts and learn how leading organizations are turning enterprise context into real business impact.
Learn how context‑aware, connected AI drives impact and usage across your company.
Walk away with a vision for how an organization can feel the value of AI on day one.
Discover the latest‑generation Glean Assistant — personalized, proactive, and a true domain expert.
IN THE KNOW
What’s trending on socials and headlines
Meme of the day
System Roadmap: System design is becoming one of the most sought-after engineering skills. This step-by-step roadmap covers everything from core concepts to architectural patterns.
One-Prompt Terminal: A single prompt turned Kimi Agent into a retro-style stock dashboard with live prices and a news feed. Full prompt included.
Claude Skills 101: An Anthropic engineer wrote a detailed explainer on how Claude Code Skills work. A must-read for devs who want Claude to follow their exact workflows.
CLI Design: A designer at Vercel shared a playbook on ditching Figma for Claude Code (from parallel sessions and git worktrees to UI tips and MCP integrations).
Code Cheat Sheet: This one-page Claude Code guide covers everything from setup to prompting techniques.
INSIGHT
How context graphs became enterprise AI's trillion-dollar idea (and how to build yours)
The architecture behind context graphs
A new concept is taking over enterprise AI. Foundation Capital recently published a blog post referring to context graphs as a “trillion-dollar opportunity.” Weeks later, Glean CEO Arvind Jain wrote about why context graphs matter. The elephant in the room? How do you actually build a context graph?
The problem. AI agents fail at roughly 70% of enterprise tasks. Separately, MIT found that 95% of AI pilots deliver zero P&L impact. The reason is the same every time: agents can reason and write code, but they don't know how your team actually gets work done.
Context graphs might be the solution. The idea is to model the recurring steps, approvals, and exceptions behind real work across your tools and org. Jain put it simply: "You can’t reliably capture the why; you can capture the how." The “how” leaves a digital trail, and over many cycles, those traces start to reflect intent. However, no deployment has published measurable results yet.
How to get started:
Pick your team's most repeated workflow, like incident response or customer escalation.
Talk to the people who actually do it and map the real sequence of steps.
Validate where good paths diverge from the bad ones, and look for blind spots where steps happen but never get documented.
To start, you can refer to Glean's breakdown of how to build a context graph.
PRESENTED BY CUSTOMER.IO
PLG only works when messaging is triggered by what users actually do.
This buyer’s guide helps founders evaluate customer engagement platforms, with a clear look at data models, workflows, and pricing traps.
AI CODING HACK
This hack fixes Claude Code's memory problem
Context hack for Claude Code sessions.
An AI product builder at ByteDance shared a custom slash command that solves Claude Code's biggest pain point. When your context window fills up and you start a new session, Claude forgets everything. Decisions, bugs you fixed, architecture choices, all gone.
Her fix was a /handover command. The setup requires creating a file at .claude/commands/handover.md with this prompt inside it.
Look back through everything we did in this session and generate a HANDOVER.md file in the current project folder. Cover:
- What we worked on and what got done
- What worked and what didn't (including bugs and how they were fixed)
- Key decisions made and why
- Lessons learned and gotchas
- Clear next steps
- A map of important filesRunning /handover before ending a session generates the file. Starting a fresh session with the HANDOVER.md file loaded gives Claude full context on everything that happened before.
You can also make this fully automatic. Tell Claude Code, "Create a PreCompact hook that generates a handover document before auto-compaction." The hook generates a new handover document automatically every time the context window is about to compress.
Grow customers & revenue: Join companies like Google, IBM, and Datadog. Showcase your product to our 150K+ engineers and 100K+ followers on socials. Get in touch.
What did you think of today's newsletter?
You can also reply directly to this email if you have suggestions, feedback, or questions.
Until next time — The Code team