Mar 7 – 13, 2026
Claude turned security researcher this week, discovering 22 Firefox vulnerabilities including 14 high-severity flaws — the kind of result that reframes what frontier models can do beyond coding. Cursor is reportedly raising at a $50B valuation, OpenAI acquired red-teaming startup Promptfoo, Meta unveiled four custom AI chips, Variant launched a scroll-first AI design tool, and Perplexity shipped a persistent Mac mini agent called Personal Computer.
On the practical side, a former Manus backend lead made the case for dropping function calling entirely in favor of Unix-style commands — one of the most useful agent architecture posts in weeks. There's also a sharp guide to exposing your design system to LLMs so they stop hallucinating tokens, and Karpathy's AutoResearch is automating model training experiments.
The big think piece is Ethan Mollick's "The Shape of the Thing," tracing AI's arc from co-intelligence to autonomous management. An essay comparing AI coding agents to the Industrial Revolution makes the economic case, and Tonsky's argument that native development has already lost to Electron is the hidden gem provocateur of the week. If you only read one thing, it's Mollick on where all of this is heading.
Releases, launches, funding, model drops, major announcements.
Anthropic partnered with Mozilla to unleash Claude on the Firefox codebase, where it found 22 previously unknown vulnerabilities — 14 rated high-severity. The bugs ranged from use-after-free memory errors to potential sandbox escapes. It's the most compelling demonstration yet that frontier models can do meaningful security research, not just generate code.
The open-source AI testing and red-teaming platform is joining OpenAI, giving the lab in-house systematic evaluation tooling as agent deployments scale.
Claude Chat gains the ability to generate interactive charts and diagrams directly in conversation, making data exploration more visual without leaving the thread.
The AI coding editor is reportedly raising at roughly $50B — nearly double its previous valuation and a signal of how much the market values developer-facing AI tools.
Variant rethinks AI design exploration with an infinite-scroll interface and a Style Dropper that transfers visual DNA between generated layouts. Exports to HTML and React.
Meta's MTIA chip family — the 300, 400, 450, and 500 — is rolling out across its AI infrastructure, reducing reliance on external GPU suppliers.
Perplexity's always-on AI agent runs on a dedicated Mac mini, with persistent access to local files, apps, and sessions — a bet that ambient compute is the next interface.
Tools worth trying, tutorials, workflows, practitioner patterns.
A former Manus backend lead lays out why they abandoned function calling in favor of Unix-style text commands for agent orchestration. The post walks through real failure modes — schema drift, retry storms, silent type coercion — and shows how plain-text command interfaces solved each one. Essential reading for anyone designing agent tool interfaces.
Karpathy's new framework automates the grind of model training experiments — hyperparameter sweeps, ablation studies, and result analysis. Open source and immediately useful.
Cursor opens up its internal model evaluation methodology, showing how they benchmark coding models on real user tasks rather than synthetic benchmarks.
A rigorous look at how infrastructure variability — network latency, container cold starts, flaky APIs — distorts agentic coding benchmark scores. Essential context for anyone comparing agent performance numbers.
A non-technical guide to building functional agents using API connectors and no-code tools. Refreshingly focused on what regular users can actually accomplish today.
A developer dissects how Claude's generative UI works under the hood and rebuilds the pattern for terminal applications. Detailed architecture walkthrough with working code.
A practical framework for keeping AI coding agents aligned with your design system — spec files, tokenized CSS layers, and automated audits that catch hallucinated tokens before they ship. Solves the real problem of LLMs inventing design values.
Trends, opinion pieces, long-form essays, deep dives, industry direction.
Ethan Mollick traces AI's trajectory from a co-intelligence tool you prompt toward something that autonomously manages workflows, decisions, and outputs. He maps the transition across education, business, and creative work — and argues the shift is already further along than most people realize. The clearest articulation of where the agent wave is heading.
The sixth edition of a16z's consumer AI ranking reshuffles the leaderboard. ChatGPT still dominates, but vertical apps are climbing fast and retention patterns are shifting.
Data-driven analysis showing AI-native startups operate with significantly smaller teams while hitting higher revenue per headcount. Nuanced about where the headcount savings actually land.
New study quantifying AI assistants' share of global search volume — with mobile dominance accelerating the shift away from traditional search engines.
Om Malik examines SoftBank's mounting debt obligations and what they reveal about the financial fragility underneath the AI infrastructure boom.
Analysis of how AI agents acting as buyers — comparing prices, negotiating terms, executing purchases — shift market power toward consumers and reshape commerce infrastructure.
An essay arguing AI coding agents have reached the inflection point where software production gets mechanized the way textiles did in the Industrial Revolution. Makes the economic case that dramatically cheaper software changes everything downstream.
Glean uses cookieless analytics by Cloudflare to count pageviews. We don’t track you across sites or sell your data. Read the privacy policy.