Idea 11 · Agentic Desk, a working name

Sell-side depth. No sell-side.

Give it a ticker. It sets analyst agents loose to research the company the way a desk would, scraping what the APIs never show, and shows its working and what it cost.

Ticker RVT
No prompt. Just a ticker.
Reviews analyst £0.74

Its plan: score 2,400 customer reviews by quarter.

done · 3 sources
Availability analyst £0.55

Its plan: track stock and delivery windows across sites.

done · 2 sources
Filings analyst £0.58

Its plan: read the last six filings for guidance.

done · 4 sources
Rivet Athleticwear · RVT Coverage note · initiation
confidence 74% cost £1.87

Thesis: the market prices Rivet as a fading label. The reviews describe a brand quietly winning its runners back.[1]

Review sentiment has climbed for five straight quarters, led by fit and durability.[2][3]

Management expects the new flagship to lift full-price sell-through this year.[4]

Researched by one agent, checked by another. 4 sources · research, not advice
Steer by picking, not typing

A scripted excerpt: a ticker goes in, the agents plan and scrape, the note assembles with its sources and a live cost.

Confident, shallow, unsourced.

Generic AI answers about a company are confident, shallow and unsourced.

The data that matters is not in the APIs: reviews, availability, footfall, what vendors and customers actually say.

A blank chat box is the wrong tool. You should not have to know the right question to get real research.

Real coverage is expensive, conflicted, and does not reach the long tail of companies.

How it works.

01

A ticker in. No prompt.

Start with a ticker, or a private company (bonds and gilts later). That is the whole input, and you do not prompt.

02

Agents write their own plan

They read what the company is (a consumer brand, so check the reviews; a grocer, so check footfall and vendors), scrape what the APIs do not have, and research on a budget you set.

03

You steer by picking, not typing

Pre-canned moves, some priced by how much work they are (comparables, a deeper dive, a review sweep). The interface builds itself around what you are looking at.

The interface builds itself around the company.

Not a fixed dashboard, not a chat box. The same product arranges itself differently for the company in front of you. Pick one.

Rivet Athleticwear · RVT Desk built for a consumer brand
Reviews to the front Sentiment up five quarters
Availability signal In stock across most sites
Filings signal Guidance: flagship opens this year
Footfall to the front Store traffic up in the north
Vendors to the front Two suppliers renegotiating
Filings signal Third depot opens this year

An illustration. Same product, a different desk for each company: reviews lead for a consumer brand, footfall and vendors lead for a grocer.

Why this works.

One agent researches, another checks it.

Disagreements are shown, not hidden. The checking layer is the differentiator.

Every claim ships with its source, a confidence and a cost.

What each claim is built on, how sure the desk is, and what it cost to produce. Auditable research, not vibes.

It plugs into Claude over MCP.

The desk is callable from where people already work.

Built by people who worked in finance.

The workflow mirrors a real desk, not a chatbot.

The business.

Usage-priced. You pay per research run and per premium move, on a budget you control, and you see the cost before you spend. Team seats on top.

Questions.

Is this financial advice?

No. Research with receipts, sources and confidence on every claim. Decisions stay yours.

Why can't I just type a question?

You can, when you want to steer. By default it proposes the research, because knowing the right question is half the job it does for you.

Where does the data come from?

Public filings and prices, plus what only scraping and agent research reach: reviews, availability, vendor and footfall signals.

What does a run cost?

You see it before you start, and a budget caps it.

You pay for the work, not a flat seat.

All ideas