TL;DR
Anthropics own internal data shows Claude now leads 26% of the company's R&D work, a milestone that signals AI agents are crossing from demo to operational reality inside the firms building them.
What happened
- Claude Code, Anthropic's agentic coding tool, is cited as the flagship example of capable AI agents emerging in early 2026.
- 26% of Anthropic's R&D work is now led by Claude, according to figures referenced in a Manchester and London Investment Trust half-yearly report dated 18 March 2026.
- The trust's chairman describes the emergence of autonomous AI agents in 2026 as "a watershed moment, comparable in significance to the launch of ChatGPT."
- The report frames this as a material divergence event: software stocks, publishing, financials, and healthcare are all being repriced as markets reassess legacy seat-based SaaS models.
- Microsoft underperformed within the trust's portfolio during this period, an exposure the manager has since materially cut, suggesting the market is already rotating away from incumbents seen as slow to adapt.
Why it matters
- AI eating its own R&D pipeline is the clearest possible proof-of-concept: if Anthropic's internal workflows are 26% agent-led, enterprise adoption timelines across every sector compress sharply.
- Seat-based SaaS is structurally threatened: the trust explicitly flags that agentic AI is a "serious substitute to humans" in enterprise functions, accelerating the collapse of per-user licensing economics.
- Market leadership is now model-release-dependent: the report warns a single model drop can swing a company from market darling to perceived obsolescence in weeks, raising the stakes for every frontier lab's release cadence.
- Disruption is broadening beyond tech: publishing, financials, and healthcare are named as newly exposed sectors, meaning the repricing wave is no longer containable to software multiples.
- The trust's own NAV fell 1.4% in the half year, partly because legacy software exposure dragged returns even as AI infrastructure holdings held conviction, illustrating the bifurcation in real portfolio terms.
What to watch next
- Anthropic's next public disclosure on internal agent utilization: if the 26% figure rises materially by mid-2026, it validates the acceleration thesis and will pressure every enterprise software vendor to respond.
- SaaS re-rating velocity: watch whether major seat-based vendors (Salesforce, ServiceNow, Workday) accelerate agent-native pivots or face continued multiple compression as institutional investors follow the trust's lead in cutting exposure.
- Competitive model releases: the report warns single releases drive sharp share-price pivots, so any Claude, GPT, or Gemini update carrying stronger agentic benchmarks will be an immediate market catalyst.
Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.