Anthropic and Perplexity push Claude and Comet into buy-side research workflows

Anthropic and Perplexity each launched a finance-focused product in the same week, Trending Topics reports. Anthropic released ten Claude agents built for finance workflows, with Daloopa and other filings and data providers feeding the underlying data layer. Perplexity followed with a product it is calling a 'Computer for Professional Finance.'

Both pitches are aimed at the same target: the manual filings-and-data work that fills the early hours of a buy-side analyst's day. Pulling segment revenue out of a 10-Q. Building a comp set. Standardising a peer group's reporting calendar. That layer is what Daloopa already industrialised for hedge funds, and what Anthropic is now wrapping in an agent interface.

The escalation matters less than the framing. Two of the largest LLM-native companies in the world are positioning themselves as workflow tools for professional investors in the same week. That positioning, more than the underlying tech, is what shifts.

What each launch actually covers

Anthropic's ten agents, per the source, are pre-configured for specific finance tasks and draw from Daloopa's normalised filings dataset alongside other third-party feeds. The pitch is that an analyst no longer needs to copy line items from a PDF into Excel: an agent does it, formatted to the analyst's model.

Perplexity's 'Computer for Professional Finance' is described in the source as a parallel offering aimed at the same user. The source does not detail Perplexity's data partners or specific capability set.

What both have in common: they sit on top of public data. Filings, transcripts, press releases, structured market data. Neither is making a claim about expert calls, private-company research, or any source that requires compliance review.

What the data actually shows

The finance research stack splits cleanly into two layers. The first is data collection: pulling the numbers, normalising the formats, building the comp set. The second is judgment: building the thesis, framing the scenarios, writing something defensible to an investment committee.

Claude agents and Perplexity's product target layer one. Useful for the grunt work. Useless for IC.

The analyst who has to defend a long position to a PM still needs the second layer, and the second layer is built on sources the foundation models do not have: expert calls, channel checks, primary-source interviews with people inside the supply chain. None of that flows through Daloopa. None of it sits in a 10-K.

This is where the positioning gets interesting for vendors. Daloopa is now an upstream data supplier to Anthropic. That is a revenue line, but it is also a category shift: filings normalisation becomes infrastructure for someone else's product, not the product itself. The same question hits every vendor that sells public-data parsing as their core offering. If Claude can call your API, are you the product or the pipe?

Three scenarios for the next 12 months

Base case. Buy-side analysts adopt these tools for the data-collection layer and keep their existing workflow for everything that matters for IC. Expert networks, sell-side research, and specialist data providers stay intact because they sell the layer the LLMs cannot reach. Daloopa and similar vendors become more important as data suppliers but less visible as standalone brands.

Bull case for the LLM players. The agent layer absorbs more of the workflow than expected. Perplexity or Anthropic strikes data partnerships with sell-side research aggregators, transcript providers, and eventually expert-network compliance layers. The buy-side research stack consolidates around two or three foundation-model platforms. Specialist vendors get squeezed into pure infrastructure.

Bear case for the LLM players. Compliance kills it. The first time an Anthropic agent surfaces something that looks like MNPI, or hallucinates a number that gets cited in an IC memo, the legal review at a Tier 1 hedge fund pulls the plug. The product retreats to junior-analyst training and earnings-week grunt work. Senior PMs do not touch it.

We read the base case as most likely. The buy-side has been through this cycle before with Bloomberg's AQR-style tooling, with FactSet's structured workflows, with Tegus and AlphaSense on the transcript side. New tools get absorbed, not adopted wholesale.

What to ask next

For anyone covering this space, the questions worth putting to a research head or chief data officer at a buy-side firm:

• Are your analysts allowed to put proprietary research into Claude or Perplexity? What is the data-egress policy?
• For the parts of the workflow these agents target, what is your current spend, and what would replacing it actually save?
• Where does compliance sit on agent-generated outputs ending up in IC memos?
• If Daloopa is now an Anthropic supplier, does that change your direct contract with them?

Watch the next two quarters for two specific signals: which sell-side research aggregators sign data deals with Anthropic or Perplexity, and whether any Tier 1 hedge fund publicly endorses either product. Both would tell you the workflow shift is real. Their absence would tell you it is not.