Matt Marshall

@VentureBeat @Sam_Witteveen went deep on why this acquisition is the official obituary for the ChatGPT era. Read the breakdown on @VentureBeat here: venturebeat.com/technology/ope…

LangChain's CEO refused to let employees install OpenClaw on company laptops — then called it one of the most important agent projects in years. The paradox explains everything about where enterprise AI is heading. venturebeat.com/technology/ope…

OpenAI bought arguably the most 'unhinged' project in AI history. Harrison Chase (CEO, @LangChain) dropped a truth bomb on our podcast: OpenClaw succeeded because it was reckless. It did the things a $100B lab like Anthropic or OpenAI couldn't do. Now OpenAI is bringing that 'unhinged' energy inside the house. Is this the final nail in the coffin for the simple chatbot? I think so. We're entering the era of 'doing' agents that write code under the hood and don't ask for permission. Full podcast drops in a week or so, but here's Harrison's take below. 👇

In my podcast with @amasad, we go deep: • Why 99% of enterprise agents fail ("slop") • The "Harness Paradox" (be ready to delete your code) • The "Verifier Loop" hack that fixes hallucinations If you're building a Vertical AI startup, watch this: youtube.com/watch?v=7i7A-Y…

Replit is trying to build this infrastructure for program synthesis. The result? They are currently ranked as one of the fastest-growing SaaS companies on the planet (per RAMP data). They are building the plumbing.

The "Vertical AI" investment thesis is getting eaten alive. Exhibit A: Anthropic’s Claude Cowork. Built in ~10 days, yet likely outperforming bespoke tools giants spent years building. The industry treated vertical agents as "expert knowledge" rather than compute problems. 🧵

For enterprise architects, "GPU strategy" is now a routing decision more than anything else. The winners in 2026 will track the workloads used, the the models they're using, where every token runs, and why. Read my full breakdown at @VentureBeat: venturebeat.com/infrastructure…

This is more than just the chip. It’s the plumbing. @tjorgensen at Supermicro told me months ago: "The whole cluster is now the computer." Feeding the beast with data is harder than the compute itself. This deal is fundamentally about memory tiering to solve that bottleneck.

Nvidia just admitted the general-purpose GPU era is over. I unpacked Jensen Huang’s $20B Groq deal over the break. The implication for enterprise builders is massive: We are entering the age of Disaggregated Inference. Here is the four-front war defining the 2026 AI stack 🧵👇

@OfficialLoganK @GoogleAIStudio Yes please!

Big upgrade to vibe coding in @GoogleAIStudio lands in Jan, but if you want to test early… 👇🏻

The lesson for AI builders: Stop chasing flashy demos. Start building pipelines that distill intelligence into tools you can actually afford to run. Build the tool, then build the agent. My exclusive deep dive on the LinkedIn "Cookbook" and their move to GPUs: venturebeat.com/ai/inside-link…

Scale breaks everything. To serve 1.3B users, they couldn't use foundation models. They used extreme distillation: • Started with a 7B policy model • Distilled to 1.7B • Pruned the student model to a tiny 220M parameters The result: A 10x throughput gain with <1% relevance loss.

The industry obsession with AI Agents is creating expensive enterprise slop. @eberger45 and the @LinkedIn team refused to build an agent until they fixed the underlying people search tool. It took months of brutal engineering. Here's how they did it. ⬇

I went deep on this shift for @VentureBeat. Spoke with Val Bercovici about why this agentic swarm revolution has turned him from a skeptic about AGI into a believer, and @MarkRuddock about building a production-ready app on a transatlantic flight. We break down the 5 architectural shifts driving this, the SWE-bench data, and the "reality check" from folks like @JeffDean. Full analysis here: venturebeat.com/ai/vibe-coding…

The conversation around AI coding is stuck on "vibe coding." There's a massive perception gap. The real story this summer is the shift to a new paradigm: agentic swarm coding. It's a move from passive prompting to disciplined orchestration, giving enterprise devs the control they want. The result? A 10x leap in autonomy and production-ready apps built on 6-hour flights.