Thesis Lab now pauses between research planning and panel recruitment so an analyst can review, edit, or delete each proposed Channel Check and Expert Call group before the AI hires anyone. The previous flow recruited automatically; the new flow asks first.
The change matters for two reasons. Recruitment costs real money - every Channel Check averages eight to ten operator interviews, every Expert Call lane fields a recruited panel - so an analyst who can prune the plan before it runs spends budget only on the angles that fit the thesis. And the plan itself is now legible: an analyst can see exactly which audiences the AI proposes, in plain English, and intervene before the work starts rather than auditing it after the fact.
What you see, in order
Once a sharpened thesis activates, the research-progress card shows three user-facing steps in plain language: Read background, Build counter-case, Identify research panels. The five backend phases that previously leaked into the UI (including the splendidly opaque "Running Ditto panels") are gone. When the third step completes, the panels surface as proposed groups across the Channel Checks and Expert Calls tabs, each with its audience plan, sample questions, and a banner across the top of the tab inviting review.
A worked example from this morning's QA run on an oil-majors basket: the AI proposed eight Expert Call lanes - Geopolitical risk strategists, Large fleet procurement and logistics leaders, Energy transition strategy consultants, Former OPEC and oil-market policy specialists, Integrated energy value-chain operators, Risk and sub-integrated energy analysts, and two more. The analyst reviewed each card, dispatched the six that fit the thesis, and deleted the two that didn't. Recruitment fired on six panels; the other two never spent a cent.
A counter strip on the main view
The research-progress card now carries a single-line counter under the chip strip: X active groups, Y ready for review, Z failed. It is the one place to see, at a glance, how many panels are recruiting versus waiting for an analyst's verdict. Click the counter and you land on the right tab to act.
Why this is the right default
A purely autonomous research agent is impressive but, at the price point of a hedge-fund expert call, also slightly terrifying. The new flow keeps the AI's planning speed - eight Expert Call lanes proposed in roughly thirty seconds - but reinstates the analyst as the budget authority. The agent surfaces the plan; the analyst signs the cheque.
Earlier in the Thesis Lab rollout: the v1 beta launch on 29th April and sector and macro thesis decomposition on 13th May.
This release ships on top of the recruitment precision work from 11th May and the API and MCP layers that went GA on 23rd May. Pre-recruitment review is on by default for every customer; nothing to enable. The full Thesis Lab walkthrough is at fish.dog/thesis-lab.