In partnership with
Today on FindMeAI:
Four stories. Each one reshapes something you assumed was settled.
→ Claude Managed Agents — anyone can now build and deploy custom agents. No infrastructure. No agent loop. Just ship.
→ Meta drops Muse Spark — 9 months, ground-up rebuild, proprietary model. Meta is back in the race.
→ The money — Anthropic passed OpenAI at $30B ARR. OpenAI projects $14B in losses for 2026. Both are burning cash faster than they're making it.
→ What it means for you — and what to try this weekend.
Claude Just Made "Build an Agent" a One-Day Project
Anthropic launched Claude Managed Agents — a suite of composable APIs for building and deploying cloud-hosted agents at scale. Claude
Translation: the hardest parts of building AI agents — infrastructure, sandboxing, state management, permissions — are now handled for you.
Before this: building a production agent meant weeks of engineering. Secure execution environments. Credential management. Agent loops. Compaction logic. Tracing. All custom-built.
After this: you define the model, system prompt, tools, and skills. Anthropic handles the rest. Prototype to launch in days, not months. Claude
Atlassian is already using it to build agents directly into Jira — so customers can assign tasks to AI right from their project board. Claude
One team built a production-ready meeting prep agent in days, calling it "3x faster" than building it themselves. Claude
Three ways to build: → Claude Console — visual, no-code → Claude Code — ask it to build the agent for you → New CLI — command-line deploy, versioning, CI/CD
Pricing: Standard token rates plus $0.08 per session-hour for active runtime. Claude
Why this matters: The agent-building moat just collapsed. If you had an agent idea but lacked the infrastructure team to build it, that excuse is gone. Today.
Meta Rebuilt Its Entire AI Stack. Muse Spark Is the Result.
After Llama 4 flopped last year, Zuckerberg didn't iterate. He started over.
Meta Superintelligence Labs rebuilt their AI stack from the ground up over nine months FB, led by former Scale AI CEO Alexandr Wang.
The result: Muse Spark — their most powerful model yet, purpose-built for Meta's products. FB
What it does differently:
→ Accepts voice, text, and image inputs Axios. Snap a photo of a grocery shelf. It ranks products by protein content. → Launches multiple subagents in parallel FB — one plans, another researches, a third compares. Faster, better answers. → Achieves comparable capabilities using over an order of magnitude less compute than Llama 4 Maverick. CNBC → Shopping mode combines LLMs with user behavior data — Meta's actual competitive advantage.
The benchmarks: Artificial Analysis scored Muse Spark at 52 — behind only Gemini 3.1 Pro, GPT-5.4, and Claude Opus 4.6. Simon Willison Llama 4 scored 18.
The catch: Unlike Meta's previous open-weight models, Muse Spark is proprietary. Fortune Available on Meta AI app and website now. Rolling out to WhatsApp, Instagram, Facebook, and Ray-Ban glasses in coming weeks. Meta plans to release an open-source version eventually. Axios
Bottom line: Meta is back in the frontier model conversation. The question is whether a proprietary model locked inside Meta's ecosystem can compete with Claude and ChatGPT's open API access.
Anthropic Passed OpenAI on Revenue. Both Are Burning Billions.
The numbers dropped this week. They're staggering — and sobering.
The revenue race:
Anthropic hit $30B annualized revenue in April 2026. Up from $1B in January 2025. 30x growth in 15 months. The-ai-corner
OpenAI confirmed $2 billion in monthly revenue — roughly $24B annualized. SaaStr
Anthropic just passed OpenAI. The company that was supposed to be the safety-focused underdog lapped the most famous AI company in the world. The-ai-corner
The burn rate:
OpenAI projects $14 billion in losses for 2026. The-ai-corner The company has committed over $1 trillion to infrastructure over the next several years. European Business Magazine: Only 5.5% of ChatGPT's 900 million users pay for a subscription. The other 94.5% use it free, while OpenAI bears the compute cost. European Business Magazine
Anthropic is burning cash too — $12 billion earmarked for model training and $7 billion for inference infrastructure in 2026 alone. TECHi®
The key difference: Anthropic projects positive free cash flow by 2027. OpenAI has pushed breakeven to 2030. The-ai-corner
Why this matters for builders: Consumer scale and revenue scale are not the same thing. Anthropic has roughly 5% of ChatGPT's consumer user base. It just passed them on top-line revenue. SaaStr The lesson: enterprise customers paying $1M+ contracts beat 900 million free users. Every time.
3 Signals From This Week That Shape What Comes Next
Zoom out. Three patterns:
1 → The agent platform war is over before it started. Claude Managed Agents, Codex Skills, Accio Work, Perplexity Computer — every major player shipped an agent-building platform in the last 60 days. The infrastructure layer commoditized overnight. The value now lives in what you build on top, not how you build it.
2 → Proprietary is back. Meta went from open-source champion to proprietary model. OpenAI never left. Even Anthropic gates its most powerful model (Mythos) behind limited access. The open-source era isn't dead, but the frontier models are increasingly locked.
3 → Revenue without profit is the industry's defining contradiction. The two biggest AI companies in the world are growing at rates that would make any SaaS founder weep. And neither will be profitable this decade without fundamental cost structure changes. European Business Magazine Every tool you rely on is subsidized by venture capital and investor conviction. That's not inherently bad. But it's worth knowing.
CTA — YOUR WEEKEND
Two things to try. One thing to read.
☐ Build your first agent with Claude Managed Agents. Open the Claude Console. Define a system prompt, pick your tools, deploy. No infrastructure needed. Start with something simple — a code reviewer, a meeting prep agent, a data analyzer. Ship it today. Iterate Monday. 30 min.
☐ Try Muse Spark on meta.ai. Log in (requires Facebook/Instagram account). Test it against your current AI on the same prompt. Try a visual task — photo of a document, a product, a whiteboard. Compare. Artificial Analysis scored it behind only Gemini Pro, GPT-5.4, and Opus 4.6. Simon Willison See if you agree. 15 min.
☐ Read the revenue/burn comparison. Not to panic. To calibrate. The tools you're building on are subsidized infrastructure. Understanding the economics behind them makes you a sharper builder and a better buyer. 10 min.
The agent platform you build this weekend might be the one your company runs on next quarter. Start while it's free to experiment. →
AI Agents Are Reading Your Docs. Are You Ready?
Last month, 48% of visitors to documentation sites across Mintlify were AI agents—not humans.
Claude Code, Cursor, and other coding agents are becoming the actual customers reading your docs. And they read everything.
This changes what good documentation means. Humans skim and forgive gaps. Agents methodically check every endpoint, read every guide, and compare you against alternatives with zero fatigue.
Your docs aren't just helping users anymore—they're your product's first interview with the machines deciding whether to recommend you.
That means:
→ Clear schema markup so agents can parse your content
→ Real benchmarks, not marketing fluff
→ Open endpoints agents can actually test
→ Honest comparisons that emphasize strengths without hype
In the agentic world, documentation becomes 10x more important. Companies that make their products machine-understandable will win distribution through AI.
🐝 One Last Thing...
Know a founder still debating which AI platform to bet on? Forward this. The revenue numbers alone will change the conversation.
Reply with what you built this weekend. Best agents get featured next issue.