The Wire WeeklyEpisode 2 · April 13, 2026 · 00:07:54.57Your browser does not support the audio element.

This Week's Stories

Full Transcript

Meta shipped the first product from Zuckerberg's entire company restructure around superintelligence BEFORE OpenAI did. Think about that for a second. The company everyone assumed was behind in the agent race just lapped the supposed leader. I'm Travis Wright, this is The Wire Weekly, and if you're still tracking this space like it's a horserace between foundation models, you're missing what's actually happening.

So let's talk about Muse Spark. This is the first thing to come out of Meta Superintelligence Labs, which if you remember, Zuck created by basically blowing up his entire AI org and rebuilding it around one thesis: agents are the product, models are the substrate. And while everyone's been watching OpenAI do billion-dollar fundraising theater, Meta just quietly shipped a production agent system. Not a demo. Not a research preview. A thing people can actually use.

Here's what matters. This isn't about whether Muse Spark is better than GPT-5.4 or Claude whatever. It's about velocity. Meta reorganized their entire multibillion-dollar AI operation, and they're ALREADY shipping. That tells you two things. One, they started this pivot earlier than anyone realized. And two, they're not treating agents like a feature bolt-on. They rebuilt the whole damn assembly line.

Now contrast that with OpenAI. They just raised 122 BILLION dollars at an 852 billion dollar valuation. Put them right up there with Walmart and Samsung in market cap. Sounds great until you read the fine print. They're planning to LOSE 111 billion dollars over the next five years. Not spend. Lose. That's averaging more annual losses than any company in history. And yeah, I get it, you gotta spend money to build AGI or whatever. But when your competitor ships first after a complete reorg and you're burning that kind of cash... that's not a lead, that's a very expensive treadmill.

And here's the kicker. OpenAI just announced a safety fellowship program. They're paying fellows 111 grand over five months, plus about fifteen thousand a month in compute credits. Which, fine, great, safety matters. Except the timing is HILARIOUS. You raise the biggest round in tech history, everyone's asking questions about your burn rate and your safety record, and suddenly you've got a splashy program paying people in GPU credits to research alignment. I'm not saying it's purely PR. I'm just saying the timing is doing a lot of heavy lifting.

But here's the thing. While we're all watching the OpenAI and Meta show, three things happened this week that nobody's connecting. And they should be. Because together, they tell you exactly where this is going.

First. Cloudflare just handed OpenAI the keys to production-grade agent infrastructure. Cloudflare's Agent Cloud now runs GPT-5.4 and Codex. This isn't a research partnership. This is enterprise deployment at scale. Cloudflare is literally building the highway for AI agents to run on, and OpenAI is the anchor tenant.

Second. SaaS stocks absolutely cratered this week. Not because of macro fears or rate cuts or whatever. Because Anthropic's Claude Cowork, OpenClaw, which got 150,000 GitHub stars in DAYS, and Google's Antigravity are now automating everything from legal contract review to full software builds. The market looked at these agent releases and went, oh, we don't need that many SaaS seats anymore, do we?

Third. Anthropic just outbid bitcoin miners for multi-gigawatt power supply. They signed a compute deal with Google and Broadcom for next-gen TPUs that requires gigawatt-scale electricity. Bitcoin miners have spent a decade cornering the market on cheap power. Anthropic just walked in and said cool, we'll pay more.

You see the pattern? Infrastructure is getting built. Real automation is shipping. And the capital, both financial and electrical, is flowing to agents, not protocols. This isn't three separate stories. This is the agent economy eating the world in real time.

And if you want proof this isn't just a Valley thing, Alibaba just wrote a 300 million dollar check for ShengShu Technology, a Chinese AI video startup most people have never heard of. That's Alibaba Cloud leading a 293 million dollar round. Not participating. Leading. China's not waiting around to see if video generation is a thing. They've already decided, and they're flooding the zone.

Meanwhile, Eclipse, the VC firm that backed Cerebras, just raised 1.3 billion dollars for a fund targeting AI infrastructure, manufacturing, robotics, and defense. Not software. Not SaaS. Not tokens. Things you can TOUCH. They're betting the next wave isn't about better chatbots, it's about agents that control physical systems. And given that Cerebras is now one of the key compute providers for anyone not going with Nvidia, I'd say they've got pretty good pattern recognition.

Okay. Wild card time. Anthropic built an AI hacker that's so damn good at finding vulnerabilities, they won't release it. Claude Mythos Preview can autonomously find deep security holes in code and build exploits to gain admin access. And Anthropic looked at it and said nope, not putting this one in the API.

Now, I love this for two reasons. One, it's the most honest thing I've seen from a frontier lab in months. They didn't ship it with a safety disclaimer and cross their fingers. They just said this is too dangerous right now, we're keeping it internal. Respect.

But two, and this is the fun part, it PROVES that these models are already capable of expert-level offensive security work. Which means every other lab has something similar, they just haven't told you about it yet. So when people ask me, are we close to agents that can do real software engineering work? I'm like, buddy, we're past that. We're at the point where labs are actively HIDING capabilities because they're worried about what happens when you put them in the wild.

That's not hype. That's the opposite of hype. That's oh shit, this works TOO well.

So what should you actually be watching? Three things.

One. Power deals. I'm serious. The Anthropic multi-gigawatt compute deal is a leading indicator. Whoever's signing electricity contracts at scale in the next six months is telling you who's actually training the next generation of models. Watch the power, not the press releases.

Two. Agent infrastructure companies. Cloudflare's Agent Cloud isn't the only one building this, but they're the first with a major foundation model partner. Whoever builds the rails for agent deployment at enterprise scale is going to print money. This is the AWS moment for agents, and it's happening right now.

Three. The SaaS replacement cycle. Those stock selloffs aren't a blip. Start tracking which categories get hit hardest. That's your map for where agents are ACTUALLY working versus where they're still vaporware. Legal tech and dev tools got hammered this week. That's not a coincidence. That's the market saying those jobs are already getting automated.

We're not in the research phase anymore. We're not even in the early adopter phase. We're in the part where the money and the power and the infrastructure all start pointing in the same direction. And this week, they all pointed at agents.

See you next Tuesday.