Two weeks ago Google spent three days in Las Vegas telling everyone this was the agentic era. The keynote was the agenda. What's interesting is how fast the rest of the industry stopped agreeing in the abstract and started carving out specific territory.
Anthropic planted a flag on Wall Street. OpenAI counter-launched in cybersecurity. Microsoft and Notion both shipped "control planes" for agents — from opposite ends of the stack. And underneath, the plumbing kept converging: Claude Platform landed on AWS, A2A and MCP spread further, and Anthropic gave its agents the ability to dream.
Here's the map two weeks later.
Anthropic Plants a Flag in Finance
The biggest move of the two weeks. On May 5, Anthropic announced a $1.5 billion joint venture with Blackstone ($300M), Hellman & Friedman ($300M), and Goldman Sachs ($150M), with Anthropic itself contributing $300M and Apollo, General Atlantic, Leonard Green, GIC, and Sequoia rounding it out. The new entity is positioned explicitly as financial services infrastructure — not "AI for banks," but the operating layer underneath them.
What ships with it: roughly ten pre-built agents covering pitchbooks, earnings analysis, credit memos, KYC, underwriting, month-end close, statement audits, and insurance claims. Full Microsoft 365 integration — Excel, PowerPoint, and Word are generally available; Outlook is in beta. Moody's embedded as a native Claude app with credit and risk data for 600M+ companies. Named customers already running it: JPMorgan, Goldman, Citi, AIG, Visa.
The Jamie Dimon quote in the Fortune piece is the one worth keeping. He asked Claude about asset swaps and Treasury bid-ask spreads, and it built him a working dashboard in twenty minutes. Dario Amodei described Anthropic's growth as 80x annualized against a 10x projection — "the cone is even wider than I thought."
That's the headline number, but the more interesting one is the strategic posture. Anthropic isn't selling Claude to banks. It's building the deployment layer banks use to run Claude. Different game.
Pair this with Anthropic's legal play earlier in the same window — 20+ new legal MCP connectors and 12 practice-area plugins — and a pattern shows up. Anthropic is going vertical-by-vertical: platform plus pre-built agents plus deep industry data partnerships. Not horizontal model competition. Specific territory.
OpenAI Counters in Security
Within days, OpenAI launched Daybreak — its cybersecurity initiative — as a direct counter to Anthropic's Claude Mythos and Project Glasswing.
Daybreak's pitch is "secure-by-construction": defense embedded into the development lifecycle, not bolted on after a vulnerability scan. It runs on a tiered model stack — GPT-5.5 for general work, GPT-5.5 with Trusted Access for Cyber for defensive workflows (code review, vulnerability assessment, malware analysis, patch validation), and GPT-5.5-Cyber for offensive work like penetration testing and red teaming.
Launch partners: Cloudflare, Cisco, CrowdStrike, Palo Alto Networks, Oracle, Akamai. That's the security-infrastructure tier of the enterprise, and getting six of them on the launch slide matters more than the model details. OpenAI is doing what Anthropic did in finance — going broad on partnerships in a single vertical rather than spreading thin.
Worth flagging the framing difference. Anthropic's Mythos got attention by finding 271 Firefox vulnerabilities for Mozilla. Daybreak's framing is preventing them in the first place. Reactive vs. proactive is a real positioning gap, but it's also marketing — most enterprises will need both. Whichever vendor gets the development-lifecycle hooks wired in deeper will win. That's an integration race, not a model race.
Two Bets on the Control Plane: Microsoft and Notion
On May 1, Microsoft Agent 365 hit general availability. $15 per user, included in Microsoft 365 E7. Three pillars: observe, govern, secure. The pitch is real-time visibility into your agent fleet — which agents are doing what, which are misbehaving, which are accessing sensitive data — across Microsoft and third-party agents alike. IT and security teams are the buyer.
That's the top-down bet. The enterprise governs agent sprawl the same way it governs identity and endpoint sprawl.
On May 13 — yesterday — Notion took the opposite end of the same problem. Their new Developer Platform launched with Workers (sandboxed code execution inside Notion), Database Sync (any API-backed source flowing into Notion databases), Agent Tools (custom logic where MCP isn't enough), and an External Agent API that lets users chat with and assign work to Claude Code, Cursor, Codex, and Decagon from inside the workspace. Since Custom Agents launched in February, Notion customers have built over a million of them.
Ivan Zhao's framing: "Any data, any tool, any agent — that's the big picture."
These are two very different bets on the same question. Microsoft is building the layer where you govern agents — IT-led, top-down, compliance-driven. Notion is building the layer where you use them — knowledge worker-led, bottom-up, productivity-driven. Both are calling themselves a control plane. They're not actually competing yet, because they're aiming at different buyers. But the logic of every platform play is that one eventually subsumes the other.
For builders, the interesting question is which integration path you wire into first. If your agentic product needs to live inside an enterprise's existing workflow, Notion's External Agent API is the easier on-ramp. If your product needs to be visible to the people who approve enterprise purchases, Agent 365's governance hooks matter.
The Plumbing Keeps Converging
Three quieter announcements in the same window that all point in the same direction.
Claude Platform on AWS (May 11). Anthropic's native platform, running through AWS billing and IAM, on Anthropic-managed infrastructure. The full Messages API, Files API, Batches, Managed Agents, Agent Skills, code execution, tool use. This is distinct from Claude on Bedrock — where data stays inside AWS. Claude Platform on AWS routes data through Anthropic but keeps your auth and billing in AWS. It's a vendor-neutrality play on the buyer side: you get Anthropic's full feature set without standing up a separate Anthropic procurement relationship.
Anthropic dreaming (May 6). Claude Managed Agents can now run a scheduled "dreaming" process that reviews past sessions, extracts patterns, and curates memory into plain-text playbooks for future runs. It doesn't update model weights. Harvey, running it in production for legal work, reported ~6x completion-rate improvement. The interesting framing isn't the magic; it's that Anthropic is now shipping agent infrastructure — memory, orchestration, outcomes — at the same cadence Google ships ADK features.
Claude Opus 4.7 (May 6). Better software engineering on hard tasks, vision up to ~3.75 megapixels on the long edge (3x prior), state-of-the-art on GDPval-AA and finance benchmarks. Pricing unchanged at $5 in / $25 out per million tokens. The first major model release framed as much around taste — "higher-quality interfaces, slides, and docs" — as around raw benchmarks.
A2A and MCP show up under all of the above. They aren't headline announcements anymore. They're the substrate.
What This Means for Builders
Three things to do with this map.
Don't pick a vertical bet too early. Anthropic owning finance and OpenAI owning security doesn't mean you have to choose a side. The vertical platforms are competing for the buyer; your job is to stay portable on the protocol layer where they all agree.
Anchor on MCP and A2A. Every announcement in the last two weeks either added MCP support, added A2A support, or did both. If your product speaks those protocols, you get to ride whoever wins the vertical fights without having to predict them.
Watch the Opus 4.7 "taste" claim. This is the first major model release positioned around professional output quality rather than benchmark numbers. If it holds up in real use, expect every other lab to follow — and expect "does it produce a slide deck you'd actually present" to become a real eval axis. That's a new kind of model selection criterion, and it favors whoever has the better designer-quality training signal.
The agentic era stopped being a slogan in the past two weeks. It started being a map. Worth knowing where the lines are before you ship.
Sources: Anthropic deepens push into Wall Street (Fortune), Introducing Claude Opus 4.7, Daybreak is OpenAI's response to Claude Mythos (Engadget), Notion just turned its workspace into a hub for AI agents (TechCrunch), Microsoft Agent 365 GA, Anthropic introduces dreaming (VentureBeat), Claude Platform on AWS
