The bar for an item here is deliberately high: real adoption by the industry itself.
Model-vendor news — a new frontier model, a benchmark, a funding round — never appears,
however loud the week. When a period produces nothing that clears the bar, the page says so
rather than padding. For the standing assessment of a given job — what the AI does, how good
the evidence is, and what it can't do — see the use-case library.
As at 2026-09-01 · Source: the trade-press or newsroom link on each item (dated per item) ·
Limitations: coverage is what crosses the trade press and managers' own newsrooms — private
deployments that are never announced are, by definition, not captured.
Recent deployments
An honest note: as at 2026-09-01, nothing new has cleared this page's bar — the items below remain the most recent real deployments to do so (adoption by the industry, not model news).
Funds Society · 2026-06-11
Janus Henderson is developing two AI-native tools with Anthropic and General
Catalyst-backed Percepta: Prism, a client-intelligence platform for distribution teams,
and Libros, a research manager that joins proprietary research, external research and
market data for analysts — alongside firm-wide deployment of Claude Code and Cowork.
The notable part is the build-with-a-partner model on top of a frontier lab's stack,
rather than buying a vendor point solution: the manager keeps the data foundation,
which is where the durable advantage sits.
Mercer · 2026-05-30
Mercer's survey of global asset managers found 55% have integrated AI into at least
one investment process and another 27% are running pilots — but, in the words of
Mercer's global manager research lead, the technology remains "a partner rather than a
decision-maker". That framing matters for governance: a partner produces inputs a human
signs off, which is a very different model-risk and accountability profile from
delegated decisions. Expect due-diligence questionnaires to start asking which of the
two you actually run.
How to read an adoption item
A deployment announcement is a claim, not a control test. Three questions separate the
ones worth acting on: who confirmed it — the deployer, the vendor, or a
journalist paraphrasing a press release; what is actually in production versus
piloted or "being developed"; and where the human sign-off sits, because a tool
that drafts is a different governance question from one that decides. The use-case library
grades exactly these, job by job. The gotcha: the firms with the most interesting deployments
are usually the least likely to announce them, so read the absence of news as missing coverage
rather than as an absence of adoption.