Anthropic's Chief Product Officer, Mike Krieger, recently confirmed a striking reality: Claude is effectively writing 100% of its own code — and much of the company's other products. In interviews and onstage discussions (including at Cisco's AI Summit in early February 2026), Krieger stated that "Claude is now writing Claude," with the model serving as the primary author for its development and related features. He referenced CEO Dario Amodei's prediction from a year earlier — that AI would write 90% of code — noting that "today it's effectively 100%."
This isn't exaggeration. Internal reports and statements from Anthropic engineers (such as Boris Cherny, head of Claude Code) show:
- For some teams, like the Claude Code group itself, AI generates 90-95% of merged code lines, with Cherny personally reporting 100% of his recent contributions (over months) coming from Claude Code and models like Opus 4.5 — zero manual edits in many cases.
- Company-wide, the figure hovers between 70-90% for code generation, per Anthropic spokespeople and internal usage data.
- Products and features (e.g., Claude Cowork prototypes) have been built in days with 100% AI-generated code in controlled internal cases.
A year ago, Amodei's forecast drew heavy skepticism — accusations of hype or dishonesty flooded comments. Today, it's not only proven at Anthropic but echoed in FAANG-adjacent circles and high-performing tech teams. Many successful engineering organizations now report 90%+ AI-assisted code (though "100%" varies by team and definition — humans still review, architect, and handle edge cases).
This is the tip of a much larger transformation. Code is verifiable and measurable — you can diff commits, run tests, and quantify automation. Most other office functions are fuzzier: harder to audit perfectly, more context-dependent, and thus slower to fully automate. But the trajectory is clear.
The Next Wave: 50%+ Automation in Non-Engineering Roles
By late 2026–2027, many white-collar knowledge tasks are projected to reach ~50% reasonable automation with current and near-term agentic AI (multi-step reasoning, tool use, memory, self-correction loops).
Examples from recent trends and reports:
- Strategy & Product Management — AI already drafts roadmaps, analyzes user feedback at scale, runs competitive teardowns, and generates OKRs. Humans set direction and make final trade-offs.
- Sales — Tools automate lead scoring, personalized outreach (email/LinkedIn sequences), CRM updates, forecasting, and objection-handling scripts. Reps focus on relationship-building and complex negotiations. Studies show sales teams can handle 3–4× more leads with AI, with 60–70% of admin tasks automated.
- Legal — Contract review, NDA triage, compliance checks, clause extraction, and basic drafting are already 50–80% automated in forward-leaning firms. Agentic plugins (like Anthropic's Legal Plugin for Claude Cowork) handle routine work; lawyers handle judgment, negotiation, and liability.
- Marketing — Content generation, A/B testing, ad copy, social scheduling, audience segmentation, and performance analysis are heavily AI-driven. Creative strategy and brand voice remain human-led.
- Support & Customer Success — Tier-1 tickets, FAQs, troubleshooting flows, and churn prediction are largely automated. Human escalation for empathy-heavy or novel cases.
- - **Finance & Operations — Forecasting, reconciliation, invoice processing, budget variance analysis, and logistics optimization are seeing rapid gains via agentic workflows.
Why ~50%? These roles involve higher ambiguity, interpersonal nuance, ethical judgment, and accountability. Full replacement is unlikely soon — but augmentation lets one person do the work of several.
Also read:
- The 73% Collapse: How AI Is Erasing Entry-Level Tech Jobs and Rewriting the Career Ladder
- All Roads Lead to Fanvue: Why AI Influencers Are Flocking to the Platform in 2026
- Why Is Fertility in South Korea So Low?
- The Egg That Hatched "Living" Robots: Tamagotchi at 30
The Stakes in 2026
- For individuals — Those who refuse to adapt face stagnation or displacement. Those who master AI tools (prompting, agent orchestration, verification loops) can 2–5× output, take on bigger scope, and accelerate careers dramatically.
- For companies — Laggards risk "gniение" — slow rot and market share erosion. Leaders who embrace "hiring hundreds of geniuses in the data center for ~$1k/month" (via API access to frontier models) gain asymmetric advantage: faster iteration, lower costs, new market capture. Anthropic itself is a living example — building products at speeds previously impossible.
This isn't sci-fi. It's happening now in the most advanced labs and scaling outward. Code was the canary; the rest of knowledge work is next. The question isn't whether automation will arrive — it's whether you'll lead it or be left behind.
