Anthropic is having the kind of year that rewrites the leaderboard. According to TechCrunch AI, the company is reportedly raising a round that would value it near $950 billion, edging past OpenAI’s $854 billion March mark, and a recent report shows Anthropic has quadrupled its enterprise market share since May 2025, now outpacing OpenAI among business customers. In a sit-down at the Code with Claude conference in San Francisco, TechCrunch AI got Cat Wu, head of product for Claude Code and Cowork, to lay out where the company is steering next.
The headline insight: proactivity is the next frontier. Wu told TechCrunch AI that 2025 was the year of synchronous development, 2026 is shaping up as the year of routines and automations, and the step after that is Claude understanding what you work on and setting up the automations for you, before you ask.
What stands out here is the shift in product philosophy. Claude is being designed to anticipate, not respond.
Why the proactivity pivot matters
Reactive AI waits for a prompt. Proactive AI watches your workflow, spots the tedious bits, and quietly handles them. That changes the buying conversation for enterprises in a few ways:
- The unit of value moves from “answers per query” to “hours of work removed per week.”
- IT and security teams get a new surface to govern: agents acting without explicit human triggers.
- Procurement starts comparing AI on observability and audit trails, not just benchmark scores.
This is significant because the competitive moat stops being model quality alone. It becomes context depth. Whichever vendor sees the most of your actual work wins.
On managing fleets of agents
Wu pushed back gently on the assumption that agents will replace expertise. Her line to TechCrunch AI: you cannot manage agents if you cannot do the job yourself. Debugging an agent looks a lot like managing a person. Did it misread the instruction? Was the request under-specified? Did the context window miss something?
That framing matters for anyone planning headcount. Wu’s pitch is leverage, not replacement. Everyone ships more, the tedious 20% of every job gets eaten by agents, humans focus on judgment and creative work. Whether that holds in practice is the open question, since the same logic was used during every prior automation wave and the headcount story usually ended up more complicated.
The competitor question
Asked how much of product strategy is reactive to rivals, Wu was blunt. Anthropic does not design against competitors. The team designs to “stay on the exponential,” meaning bet that models keep improving and ship to the frontier of what they can do. If you chase competitors, she said, you end up perpetually a month behind.
It’s a confident posture, and it lines up with the release cadence. At least six models last year, nearly as many already this year, plus restricted releases like Glasswing, the cybersecurity model Mythos shared only with a small consortium including Amazon, Apple, CrowdStrike, and Microsoft because Anthropic deemed it too dangerous for public release.
Practical takeaways for operators
If you build with or buy AI, three moves make sense now:
- Audit your team’s tedious work. Email triage, ticket routing, status updates, report generation. That backlog is the first thing proactive agents will eat. Map it before vendors map it for you.
- Build agent management as a skill. Treat it like hiring a junior. Write specs, review outputs, debug failures. Companies that develop this muscle in 2026 will run circles over those still prompting one-shot.
- Decide your governance posture before agents start acting unprompted. Logging, kill switches, scope limits. Do this while the agents are still asking permission.
What to watch next
Wu’s prediction has a clear track record behind it. She joined Anthropic in August 2024 and helped take Claude from chatbot to coding tool to agent platform in roughly 18 months. If the proactivity wave arrives on her timeline, the AI buying cycle in 2026 will look very different from the one happening now. The full interview is worth reading at the original source.