The model isn't the bottleneck anymore. Tech & AI Weekly by Eli, July 24 2026
Hi! I spent the week in NYC between an onsite and a video studio, and the news cycle did not slow down for me. Here’s what actually mattered, curated so you don’t have to follow 20 sources.
AI & GenAI
-
The bottleneck in AI agents is no longer the model. It’s the context layer: what the agent knows, remembers, and retrieves. Good breakdown of why your agent architecture matters more than your model choice. Read more
-
Amazon, Microsoft and Google are converging on the same enterprise agent architecture. Three companies that agree on nothing now ship near-identical agent platforms. The catch: you still can’t move an agent between clouds. The standard is forming, portability isn’t. Read more
-
A judge approved Anthropic’s $1.5B copyright settlement. About $3,000 per book to thousands of authors whose pirated copies trained Claude. Training data now has a court-approved price tag, and every AI lab is doing that math this week. Read more
-
Kimi K3 topped the coding leaderboards, then demand broke it. Moonshot’s open-weight model got so much traffic they had to pause subscriptions in 48 hours. The interesting part isn’t the benchmark, it’s that inference capacity is now the moat. Weights are free; GPUs are not. Read more
-
InfoQ interviewed Clare Liguori on Strands Agents. The technical lead of the open source SDK on the design decisions behind it, straight from the source. If you’re building production agents, worth your time. Read more
Cloud & Dev tools
-
GitHub Copilot vs raw API access: what are you actually paying for? GitHub’s own math on the abstraction layer. Useful before your next “we could just call the API directly” meeting. Read more
-
GitHub Sponsors crossed $100M paid to open source maintainers. The first $10M took two years. The latest $10M took five months. Funding maintainers is finally picking up speed, and the gap is still enormous. Read more
From me this week
- AI Agent Memory Types: Your Agent Forgets Everything. Fix It. New post. Your agent’s amnesia isn’t a broken model, models are stateless by design: memory belongs to the harness you build around them. I map the four places memory usually lives (key-value, vector, graph, hybrid), what each costs, and how the same stored fact can be findable or invisible depending on the way in. Read more
I hope you enjoy this as much as I enjoy putting it together every week. It’s my little weekly de-stress moment. Let’s stay informed together and learn something new along the way.
See you next Friday,
Eli
Dev.to · LinkedIn · GitHub · Twitter/X · Instagram · YouTube