Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
0:00
|
7:30
The CEO of Google DeepMind says the way AI took off wasn't what he intended — and his alternative vision might change how you think about everything happening right now. This weekend also surfaced a story about an AI that ran a completely autonomous operation for one hundred straight days with zero human input, and the results made TIME Magazine. And two of the most powerful tech companies in the world are in crisis mode over their AI products, for very different reasons. Catch up on everything you missed this weekend in under 10 minutes.
Feedback? Email mike@yesterdayinai.news or connect on LinkedIn, X, Bluesky, or Substack. If you like the show, please take a minute to rate and review it so others can find it!
Hi folks, this is Yesterday in AI, your daily digest of everything happening in the world of artificial intelligence. I'm Mike Robinson. It's Monday, April 13th, and this weekend the AI conversation went somewhere genuinely philosophical. The CEO of Google DeepMind admitted that the way AI took off wasn't exactly how we wanted it to go. We'll dig into that, plus the story of an AI that quietly grew a tomato for 100 days, Slack becoming something entirely different, and Microsoft going into what employees are calling a code red. Let's get into it. We'll start with Demesis Sabas, because his comments from a new interview published this weekend deserve more than a quick headline. In a conversation with journalist Cleo Abram, the CEO of Google DeepMind made a striking admission. The chatbot revolution wasn't what he had in mind. In his ideal timeline, AI would have stayed in research mode longer, quietly solving hard problems in medicine and energy, running behind the scenes the way Alpha Fold did, before any consumer product launched. He described a CERN-like collaborative model where the most capable AI was built carefully, under shared global oversight, rather than in a race fueled by commercial pressure and geopolitics. What happened instead is history. Transformers turned out to work better than almost anyone expected at generating coherent language. OpenAI shipped ChatGPT, it broke every record, and the entire field got pulled into what Hasabas calls a ferocious commercial pressure race. He's not saying the chatbot era was worthless. He acknowledges that public access accelerated progress and that millions of users stress testing these systems in the wild taught researchers things they couldn't have learned any other way. But it's clear that when he imagines the most important work AI can do, he's not thinking about meeting summaries or browser assistance. He's thinking about Alpha Fold, which he says now touches the work of more than three million scientists and may influence nearly every drug developed in the coming decade. This matters not just as a personal biography from a major lab founder, but because of what it says about where the real ambition in this field lives. While the rest of the industry fights over benchmarks and subscription tiers, Hasabas is describing a different goal. One where the chatbot moment we're living through is, at best, a useful detour. Now for a story that perfectly captures both sides of that tension. An experiment that started as a hobby project and ended up on the cover of Time magazine. A developer named Martin DeVito spent 100 days completely hands-off on a grow tent in his home. He wired up soil sensors, temperature probes, a light meter, and an Arduino microcontroller, gave Claude full control of the equipment, and planted a tomato seed. Then he stepped back. Claude, running continuously with a system called Soul, made every single growing decision for 100 days. When to water, how much CO2 to introduce, when to cycle the fans, how to adjust the light spectrum as the plant matured. No human touched a dial. At the end of a hundred days, the tomato plant, which the project calls Soul, had survived, thrived, and fruited. The experiment went viral online, generating tens of millions of views, and now has a space payload company reaching out about growing experiments off Earth. What strikes me about this project isn't just the technical achievement. It's the proof of concept for AI systems that can hold a real-world feedback loop open indefinitely, adapting in real time to something as unpredictable as a living plant. That's a very different category of task than drafting an email. Moving on to Slack, which ships something this weekend that's a lot bigger than most people realize. The new Slackbot, they've rolled out more than 30 new AI features, is now operating as a full context aware agent, powered by Claude. The core idea is that Slackbot already knows your work, it reads your channels, files, and past conversations, and it uses all of that context before you've asked it anything. You don't have to re-explain your team, your project, or the decision history that got you here. It just knows. Slackbot now functions as an MCP client, coordinating across Google Workspace, Microsoft 365, Notion, Workday, and ServiceNow through natural language requests. But the thing sticking with users is simpler, reusable skills. You define a workflow once, a pre-meeting briefing, a post-call action item summary, and it runs with a single command for anyone on the team. Deepview reports team saving up to 90 minutes per person per day. When AI is embedded inside the tool where work already happens, the friction drops to almost nothing. This is a meaningful platform shift, and it's happening largely without a press conference. Over at Microsoft, something is going badly enough that CEO Satya Nadella apparently used the phrase code red. That's the internal label being applied to an emergency overhaul of Copilot, Microsoft's AI productivity suite that has, by most accounts, underperformed since launch. Investor confidence has wobbled. Copilot hasn't moved adoption metrics the way the company promised, even as Microsoft has poured billions into OpenAI and built AI capabilities across its entire product line. The Code Red campaign is pushing teams to ship improvements faster, with a new E7 suite launching May 1st and further updates rolling throughout 2026. The competitive pressure is clear. Anthropic is increasingly winning enterprise deals that Microsoft expected to own, and the gap between what Copilot promised and what it's delivered is a real vulnerability. Azure remains strong, free cash flow is healthy, and analysts still see a rebound path if adoption numbers improve. But the fact that Microsoft is running an internal code red on its flagship AI product is a signal worth noting. Being a good AI company turns out to require more than just having access to good models. I want to spend a moment on a developer story that got some coverage Saturday and touches on something we've been watching closely. Peter Steinberger, the creator of OpenClaw, was temporarily suspended by Anthropic for what the company called suspicious activity. We covered the subscription policy change where Anthropic stopped covering third-party harnesses in last week's episode, but the creator ban adds a new layer. Steinberger posted about the suspension publicly. Anthropic restored his account within hours. But his public response was pointed. He accused Anthropic of borrowing OpenClaw's core ideas into its own agent platform while simultaneously restricting open source tools that compete with it. Worth noting that Steinberger now works at OpenAI, which has been quick to position itself as the developer-friendly alternative. This story is probably not over. A few quick items before we close. Sci-5, the RISC-V chip company founded by the original Berkeley engineers who designed the open instruction set, raised $400 million this weekend at a $3.65 billion valuation, backed by Nvidia and a mix of institutional investors. Sci-5 licenses CPU designs, it doesn't manufacture chips, and it's positioning RISC-V as an alternative to X86 and ARM for AI data center workloads. Critically, its designs will work with Nvidia's CUDA software and NVLink Fusion. The AI hardware ecosystem is broadening, and open standards are a meaningful part of the story. On the robotic side, Superhuman covered the weekend's highlights, including Clone Robotics unveiling a roadmap to bring Protoclone, their polymer skeleton humanoid, to market by 2028 at under $20,000 at scale. Sincere's Loom robot, disguised as a floor lamp that unfolds robotic arms to fold laundry and make beds, is now taking pre-orders. And Waymo launched fully driverless rides in Nashville, opening 60 square miles to public riders. And finally, Sam Altman published a rare personal post after a Molotov cocktail was thrown at his home last week. He posted the family photo, called on the industry to de-escalate, took responsibility for past conflicts, and wrote that no single company should control AGI. For the CEO of the most powerful AI lab in the world to be talking publicly about personal safety and industry tensions in the same post, that is where we are right now. That's all for this weekend catch up edition of Yesterday in AI. Stay curious, and I'll see you tomorrow.
