A Weekly Digest → The Agentic Era meets Reality
The most important shifts in AI during the week of June 1–7. It’s all about agents
What was the most interesting last week? There were two important conferences – Microsoft Build and Snowflake Summit – and we also has a conversation with Mario Rodriguez, Chief Product Officer at GitHub, about the future of developers in the agentic era. The core trend appeared to be what comes to the forefront when AI agents meet reality. Let’s unpack the core questions.
Enterprise AI Middlemen: Who Survives the Agent Era?
Agents are getting stronger faster than companies can absorb them. Reporting straight from Snowflake Summit and Microsoft Build, the question we’re asking is Will “middlemen” like Snowflake, Microsoft, Databricks, and Salesforce be replaced or become the trust layer agents need?
Stop Babysitting Agents, Start Authoring Outcomes - Guest post by Raymond Weitekamp
Great Claude Code or Codex sessions are easy to lose and hard to repeat. This article is about a useful tool – OpenProse that turns your best agent workflows into reusable .prose.md programs written in logical English, so agents can rerun them, review, and leave receipts. Not full determinism, but definitely less babysitting.
GitHub’s Mario Rodriguez on AI Coding Agents, Copilot, and the Future of Developers
GitHub was built for developers collaborating with developers. But now it needs to be reimagined for human-agent collaboration – from UI to UX to AX, where developers, new builders, and agents share the same canvas. Mario Rodriguez explains what macro-delegation is and why the future developer may simply be a builder with intent.
“GitHub’s mission is advancing human progress through developer collaboration. And now we probably need to say developer and agent collaboration,” – he says.
Reasoning RL in 2026: GRPO, DPO, RLVR, Agentic PO & Beyond
Another important topic is the evolution of optimization techniques. Reasoning RL in 2026 is getting crowded fast: GRPO, DPO, RLVR, DAPO, GSPO, ARPO, VPO, and many others. This guide maps the new toolkit for models and agents that reason, verify, search, self-correct, and improve with every optimization step – from cheaper critic-free methods to agentic and test-time training.