Precis AI recruits policy wonks for Willard's citation-first beta

Why it matters

Precis AI is testing whether founder expertise and federal-record grounding can create a defensible policy AI product before established legislative platforms absorb the same workflow.

Precis AI founder David Fuscus opened access Friday to a beta of Willard, an AI workspace built for lobbyists, government-affairs teams and congressional staff who need answers tied to federal records.

The September 18th push is less a conventional product reveal than a call for policy professionals to help shape the software. Precis AI said in its beta announcement that selected users will receive a direct line to the product team. Early access is free and invitation-only, according to Willard's website.

That customer-development strategy fits Fuscus, a longtime Washington public-relations executive who has spent much of his career close to the machinery Willard is supposed to understand. Before founding Xenophon Strategies in 2000, he served as deputy chief of staff to Pennsylvania Gov. Tom Ridge and held communications and staff roles in the House of Representatives, according to a published biography. He also worked on George H.W. Bush's 1992 reelection campaign.

Fuscus says he sketched the original Precis AI business plan on a McDonald's bag at a highway rest stop in 2022. The idea was to build AI around the vocabulary, documents and working habits of a profession, starting with public relations. Co-founder and product chief Bo Hrytsak oversees technical development, while co-founder Chris Fuscus leads prompting and the incorporation of subject-matter expertise.

Willard carries that thesis into a field David Fuscus already knows: Washington professionals trying to interpret dense public records under deadline pressure.

Grounding is the product claim

Willard retrieves information from what Precis AI describes as millions of federal and congressional documents. Its functions include plain-language policy research, legislator profiles, voting and position analysis, comparisons between bills and amendments, hearing analysis, shared document workspaces and a Highlight & Click fact-checking tool.

Precis AI says Willard coordinates 16 AI systems, using agents to assign work among models from providers including Anthropic, OpenAI, Google and Perplexity. Precis AI also says uploaded customer documents remain inside its DataVault environment and are not retained for model training.

The multi-model design is central to Fuscus' accuracy pitch. Rather than treating a single model's response as final, Willard is designed to have models check claims, identify conflicts and retrieve supporting government documents. Fuscus described the immediate problem as "a confident, cited AI generation that's just false."

That is a recognizable failure mode in Washington. An August investigation reported that AI-generated legislative drafts were increasing the workload of the House Office of Legislative Counsel, with attorneys correcting statutory language, citations and definitions that staffers and advocacy groups had produced with general-purpose chatbots. Former House Legislative Counsel Wade Ballou warned that models can miss distinctions among mechanisms such as tax credits, deductions, exclusions and grants.

Precis AI also points to the 2026 Stanford AI Index, which published hallucination rates ranging from 22% to 94% across model configurations on the AA-Omniscience benchmark. That result measures performance on broad knowledge questions and a model's willingness to give a wrong answer instead of declining, according to the benchmark research. It does not measure legislative drafting, federal-record retrieval or Willard's performance.

Precis AI has not published a comparable independent evaluation for Willard. The size and claimed primacy of its government-document database also remain company assertions. Citations make errors easier to inspect; they do not establish that the retrieval system selected the right authority, interpreted it correctly or captured a recent amendment.

A beta that began before Friday

Friday's announcement calls September 18th the opening of Willard's beta. Precis AI had already introduced Willard on June 11th as an invitation-only product, and O'Dwyer's reported at the time that Willard had been in beta testing for several weeks.

The new announcement therefore marks a broader recruitment effort for beta participants, rather than Willard's first appearance. Fuscus is explicitly offering those participants influence over how the product handles policy work. That gives Precis AI access to the corrections, edge cases and workflow knowledge needed to turn a federal-document search layer into professional software.

It also shifts part of Willard's validation into the beta. Policy professionals will determine whether its sources hold up, whether comparisons catch consequential changes and whether the multi-model fact checker reduces errors without slowing routine work.

Washington's policy platforms are adding agents

Willard enters a market where established policy-software vendors are already attaching AI interfaces and agents to large legislative datasets.

Quorum's Quincy searches federal, state and local policy information alongside an organization's meeting notes, advocacy history and stakeholder data. On September 9th, Quorum launched Quincy Hub with agents for bill tracking, meeting preparation, CRM updates, campaigns and policy communications.

FiscalNote's PolicyNote covers Congress, all 50 states and over 100 countries. FiscalNote has also expanded its API to feed legislative and regulatory intelligence into outside applications and AI-agent workflows.

Precis AI is taking a narrower route. Willard concentrates on federal and congressional records, paired with public-affairs workflows and customer-data controls. That focus gives Fuscus a credible opening with users who care deeply about provenance and spend their days inside federal material. It also leaves Willard competing against platforms with broader jurisdictional coverage, years of customer data and established tracking systems.

Precis AI's first vertical offers only a limited public guide to adoption. Friday's release says Precis Public Relations has users in 14 countries, while Precis AI's current team page says 28 countries and claims users save an average of 61% of their time per task. Precis AI has not supplied the sample, methodology or timeframe behind the savings figure, and the two geographic counts use no stated common measurement date.

Fuscus' bet remains straightforward: professional AI will be won through specialized data, workflow design and practitioners who can recognize a plausible-looking mistake. Willard's beta puts that proposition in front of a particularly unforgiving customer base. In policy work, an answer can be fluent, sourced and still wrong in the clause that matters.