Public relations

Best AI tools for a deep tech PR agency in 2026

Claude, Obsidian, Claude Cowork, Claude Code and Google Drive. Thirteen products named on the record by Sam Polstein, and only two are PR software.

A deep tech PR agency's AI stack turns out to be four general-purpose products and a backup folder: Claude for the work, Obsidian to hold the context, Claude Cowork to point the model at the right folder, Claude Code to build internal dashboards, and Google Drive for backup. Sam Polstein, who runs communications and content for deep tech startups at HAUS, named thirteen products on the record. Two of them are PR software, and both are the ones he criticizes.

Every tool below was named by Polstein in an interview with Artificial Reality, the AI buyer-evidence podcast hosted by Galina Fendikevich. Nothing here is a recommendation. It is a record of what one operator reported using, and how.

What sits at the center of the stack?

Two products, named together. Asked what matters most to how the agency operates, Polstein answered with a pair:

"I would say two things are the most important tool for how we do business, which is Claude, an enterprise-grade LLM that can help us [with] everything from diagnosing to taking a first crack at certain things."

The second, named moments later, is Obsidian. Model plus context library. That is the whole architecture.

What is Claude actually used for?

Drafting, diagnosis, crisis and rapid response — and one use he singles out as very strong that is not writing at all. He feeds the agency's own checklist and client call context back in so the model catches what was missed. His example of the output:

"Your client Kyle said in the call recording on May eighth that he wanted this thing and this was really important to him. So Sam, don't leave that out."

What he criticizes, in his own list: the recognizable house voice ("Claude has a tone"), the absence of emotional resonance, weak performance on media list building, being too agreeable to argue with, and usage limits that "definitely come into play pretty quick," particularly during dashboard work.

Why keep everything in Obsidian?

Because markdown is readable by any model and owned by him. Polstein's reason:

"For me personally I love using Obsidian as a way to keep my thoughts in one place and store them as markdown files, which are super easy for Claude to read."

What sits in it: client call transcripts, decks, technical white papers, academic research, house process documentation, and a folder he calls "good writing" holding examples plus a list of tropes to avoid. Google Drive backs the whole thing up — "at the end of the day it's just a bunch of folders of markdown files, plain text files."

Claude Cowork is how the model reaches it. In his description, Cowork can jump in and, when directed at a certain folder, see all the context for a given client or process, including the examples of good writing it then edits its own output against.

What does he build instead of buying?

Dashboards. Polstein uses Claude Code for client-facing competitor tracking and for what he calls helpful analytics in terms of GEO and answer engine optimization — reporting that in his account used to be either very expensive and a huge part of agency overhead, or did not exist. He mentions Codex alongside it as part of the same shift. The boundary he draws is firm: this is not enterprise software development, it is "making a pretty simple solution, like building a dashboard."

Which PR platforms does he still pay for?

Muck Rack and Cision, used across his career, and the subject of his sharpest professional frustration. He calls them CRMs for exactly the journalist-relationship job, then says of the data:

"That sort of information, it's still not that easy to come by, still somewhat out of date — although less so than if I was just prompting Claude or ChatGPT."

They outperform a general-purpose model on journalist data and still fall short of what the job requires. That is the only verdict he gives them.

What about the other models?

Four more are named, with decreasing evidence of use:

ChatGPT — referenced throughout, not his primary tool. He prefers Claude for being less effusive, and cites ChatGPT as what clients use to do the work themselves at roughly fifty percent quality, the floor an agency has to clear.
Perplexity — used for a period and found "a little stronger" for the things he previously took to Google.
Gemini — one line on recency, "I think maybe Gemini is a little bit better at this," with no usage claim attached.
Fable — playing around with it a little the day before, no assessment offered.

Crimson Hexagon appears only as history: a social listening platform hyped early in his career at roughly $20,000 a year for access.

What should you copy?

The order of operations. Polstein did not buy AI for public relations. He bought a model, put his own judgment in plain text next to it, and only then automated anything — which is why he can switch vendors without losing the part that took years to build.

If you run a comms function, start by writing down the process and the examples of work you consider good, in files you own, before you evaluate a single product. The stack above is unremarkable. The library is the asset. What would yours contain if you started it this week?