The thesis is simple. Anthropic's strategy is not to match OpenAI feature-for-feature. It is to build modular, enterprise-grade products (Claude Code, Cowork, Agent SDK) that embed into how work already happens. OpenAI is trying to get everyone to live inside one app. These are two different platform bets. One is shipping. The other is still in development.
If you run a business and use AI as operational leverage, the last 60 days just changed the landscape under your feet. Here is what happened, what it means, and what to do about it.
The Model Layer: February's Foundation
Start with the models, because that is where the whole run began.
- February 5: Claude Opus 4.6. Adaptive thinking. One million token context in beta. 128,000 token max output. The best reasoning model available on any platform, arrived without fanfare, without a livestream, without a countdown clock.
- February 17: Claude Sonnet 4.6. The first Sonnet-tier model to beat the previous generation's Opus at coding tasks. Same price as Sonnet 4.5. Anthropic did not charge more for a better model. They held the line and let performance do the talking.
- March 10: One million token context went GA. No surcharge. No beta headers. No special tier required. Industry first at standard pricing. And Haiku 4.5 hit $1 per million input tokens, one-third the cost of the previous Sonnet tier.
The model layer is now table stakes. Both labs are at rough parity on benchmarks. The war moved to platforms. Models are the engine. What matters now is the car.
Claude Code Becomes Infrastructure
This is where the story gets interesting. And fast.
Between late February and mid-March, Claude Code shipped nearly 20 releases (v2.1.63 through v2.1.81). Not patches. Features. The velocity was unlike anything the developer tools market has seen from a single product.
- March 3: Three releases at once. Voice mode with push-to-talk across 20 languages.
/loopfor session-scoped cron (type/loop 5m check the deployand Claude monitors every five minutes). And--worktreefor isolated git sessions on the same repo. - March 6: Local scheduled tasks in the desktop app. Set Claude to run a task at 7am every morning. It runs whether or not you are at your desk.
- March 9: Code Review. A multi-agent PR review system that dispatches a team of agents per pull request. The first real "agent team" feature in a mainstream developer tool.
- March 11: Auto Mode. Claude judges which actions are safe and handles permissions autonomously. The shift from "supervised assistant" to "trusted operator."
- March 14: MCP elicitation. Claude can pause, ask you a question through a connected service, get your answer, and continue. Interactive agents, not just autonomous ones.
- March 20:
--channelsflag. MCP push notifications to active sessions via Discord, Telegram, or any messaging platform. - March 21 (Today): Cloud scheduled tasks on the web. Close the tab. It keeps running. Claude Code stopped requiring a terminal. Schedule work, close your laptop, pick it up from your phone.
Other releases filled in the gaps: HTTP hooks for custom integrations and real-time dashboards. The --bare flag for scripted and CI calls. Remote Control (February) to start a session in your terminal and pick it up from your phone.
Claude Code went from "smart terminal assistant" to "always-on development infrastructure." You can schedule it, remote-control it, loop it on recurring tasks, dispatch teams of agents, and let it run in the cloud without a terminal open.
Cowork: Claude Code for Everyone Else
Claude Code is for developers. Cowork is Claude Code's philosophy rebuilt for everyone who is not a developer.
Launched January 12, Cowork brought persistent, agentic Claude to Pro users. By January 23, it was available on Team and Enterprise plans. The product vision: an AI coworker that remembers your context, runs tasks on a schedule, and operates across your tools.
- February 25: Scheduled tasks. Daily briefs. Weekly reports. Friday presentations. Set them once, they run on cadence.
- Mid-March: Dispatch. A persistent conversation that runs on your computer, accepts messages from your phone, and spins up sub-agents for complex tasks. A command center for your AI coworker.
The plugin marketplace: Google Drive, Gmail, DocuSign, FactSet, and a growing list of enterprise APIs. Persistent memory. Mobile access from iPhone and Android.
On March 9, Microsoft announced Copilot Cowork, built with Claude, bundled into the Microsoft 365 E7 tier. Anthropic is now embedded inside the world's largest enterprise software platform. Every E7 customer gets Claude-powered Cowork alongside Word, Excel, Teams, and Outlook. This is not a partnership announcement. This is distribution at a scale no AI startup has achieved before.
The Infrastructure Play
Products ship and die. Platforms compound. Anthropic's moves over the last 60 days look less like a product roadmap and more like a platform buildout.
| Move | What It Means |
|---|---|
| Vercept Acquisition (Feb 25) | Cloud computer-use agent with remote MacBook infrastructure. Claude can now operate full desktop environments in the cloud. |
| Bun Acquisition (Dec 2025) | JavaScript runtime now powering parts of Claude Code. Faster tooling, tighter control. |
| Agent SDK | Python + TypeScript. Same building blocks Anthropic uses internally: hooks, subagents, MCP, permission modes, session resume. |
| MCP Ecosystem | 3,000+ connectors. Elicitation, server-push channels, Tool Search GA. A functioning protocol ecosystem. |
| Partner Network | $100M committed. Snowflake, Harvey, Replit building on Claude infrastructure. |
| Series G (Feb 12) | $30B raised at $380B valuation. $14B ARR. Capital to sustain this pace. |
Anthropic is not just building products. They are building infrastructure that other companies build on. The Partner Network, the SDK, the MCP ecosystem, the Microsoft deal: these are platform moves, not product moves.
The Competitive Frame
Context matters. While Anthropic shipped 20+ releases in 60 days:
OpenAI announced a desktop superapp. It is still in development. No launch date. OpenAI's Codex remains cloud-sandboxed and asynchronous. Claude Code is local, interactive, and now also cloud-capable. Claude overtook ChatGPT as the #1 app on Apple's App Store on March 1. Claude's US business market share reached 70%, up from 10% in February 2025.
Where OpenAI still leads: multimodal capabilities (image generation, video, native voice), consumer mindshare, the Atlas browser agent, and certain terminal-specific benchmarks. These are real advantages. Dismissing them would be foolish.
But the strategic picture is clear. OpenAI is building toward a future product. Anthropic is shipping a present platform. One strategy requires users to wait. The other gives them something to use today.
For founder-operators making infrastructure decisions right now, "shipping" beats "announcing" every time.
06 · TakeawaysWhat This Means for Founder-Operators
The "which AI tool" question is now a platform decision.
Stop evaluating AI tools like SaaS point solutions. This is an infrastructure choice. If your developers use Claude Code, Cowork extends that same system to your non-technical staff. If you run on Microsoft 365, Claude is now embedded in your stack via Copilot Cowork. If you build internal tools, the Agent SDK gives you the same primitives Anthropic uses. These choices compound. They create switching costs. They lock in advantages.
This is not "which chatbot should we use." This is "which AI platform do we build on for the next three years."
Always-on AI is here. Use it.
Cloud scheduled tasks, /loop, Dispatch, remote control. Claude can now run when you are not at your desk. This is the shift from "AI as tool I use" to "AI as system that runs."
Map your recurring operational tasks. Daily reports. PR reviews. Data pulls. Inbox triage. Competitive monitoring. Weekly summaries. Start scheduling them. The founder who does this reclaims 5 to 10 hours per week. The founder who does not is still doing the work manually.
The cost of waiting just went up.
Claude Code's market share is doubling quarterly. The companies building on this infrastructure now are accumulating compound advantages in speed, institutional knowledge, and operational leverage. Every week you build on the platform, your systems get smarter. Your prompts get sharper. Your workflows get faster.
The gap between "we use AI sometimes" and "AI runs our operations" is widening every week. It was a crack in January. It is a canyon by June.
Pick your platform. Start building.
Sources: Anthropic Blog, Claude Code CHANGELOG, TechCrunch, Bloomberg, CNBC, Fortune, GeekWire, X posts from Alex Albert, Boris Cherny, Felix Rieseberg, Lydia Hallie, Thariq, Cat Wu, Dickson Tsai
Kent Langley builds Founder OS, an AI-augmented operating system for founder-operators. If you want to see this in action or build your own skill library, reach out.