Specset ✓

Specset is an AI document-control platform for construction, formerly known as Specbook AI, founded by Gordon Hempton and Wes Hather (co-founders of Outreach.io, the sales-engagement platform they helped scale to a $4 billion-plus valuation) alongside Graham Ralston, a former Senior Advisor at the U.S. Department of Transportation -- a notably credible founding team combining genuine enterprise-software scaling experience with real federal-infrastructure domain knowledge, headquartered in Seattle, WA.

The core idea is reading a project's specifications, drawings, submittals and RFIs as a single, unified record rather than as separate documents reviewed in isolation -- every finding Specset's AI agents surface is cited directly to its source page, so an engineer can click through and verify the exact spec section or sheet a conflict, omission or non-compliance was found against, rather than trusting an opaque AI summary.

That record-first architecture shows up in how Specset handles messy, real-world document sets: addenda and special provisions are merged into the base spec they amend ("conformed, not indexed"), revision history is tracked sheet by sheet, and the platform explicitly supports non-standard formats most software ignores -- Caltrans, WSDOT, FDOT and UFGS specification books, railroad and port-authority specs, and scanned legacy manuals, with the vendor stating a new format is "a normal Tuesday, not a support ticket."

Specset runs a distinct AI agent for each phase of a project's lifecycle rather than one general-purpose assistant: a design-quality-review agent cross-references every spec section against every drawing sheet at each milestone set; a bid-screening agent reads an entire solicitation the day it lands and runs a 93-point contract review; a submittal-review agent checks hundreds of individual requirements per package against the governing section and returns findings annotated directly on the submittal's own pages; an RFI agent drafts and tracks RFIs and verifies answers against the latest drawing revision; and closeout and facilities agents carry the same cited record from turnover into ongoing operations, tagging assets, warranties and maintenance schedules.

A distinctive technical detail is adversarial verification: after an agent files a finding, a second, independent agent actively tries to disprove it before it's surfaced to a human reviewer, with rejected findings archived with reasoning rather than silently dropped -- a genuinely different quality-control approach than a single-pass AI review, aimed specifically at avoiding the false-positive noise that makes engineers distrust AI-flagged issues.

The platform has real scale behind it specifically in federal and public work: the company states its record spans more than 2.6 million pages of contract documents, 90,000-plus revision-tracked drawing sheets, 31,000-plus conformed spec sections (Caltrans and UFGS included), 14,000-plus submittals managed, and over $120 million in rework exposure caught before construction, across customers including HNTB, Guardian Construction, Turnagain Marine Construction, Ghirardelli Associates and DCS Marine Services. Security posture is built for that federal/public-sector audience specifically: a completed SOC 2 Type II examination (report issued August 2026), a filed FedRAMP 20x Class A application under review, CMMC Level 1 self-assessment aligned to NIST SP 800-171, and an explicit, published commitment that customer data never trains the underlying models.

Editor’s takeBIM•STACK editor's take: Specset earns a genuine editor's pick in this catalog -- its combination of a credible, proven founding team, a cited-to-the-source-page architecture that avoids the black-box problem plaguing many AI document tools, adversarial self-verification before a finding ever reaches a human, and real production scale specifically on hard federal/public-sector document formats (UFGS, Caltrans, DOT books) is a rare, well-executed combination. Pricing is contact-sales-only with no published tiers, which is the honest cost of evaluating it, and firms outside federal/public construction should confirm the platform handles their specific document conventions as smoothly as it clearly handles government spec books -- but for GCs, CM firms and A/E firms doing that work, this is one of the more technically serious AI document-control platforms available today.

Specset's onboarding is explicitly designed to avoid the setup friction common to document-management platforms -- the vendor's own description is "upload your PDFs, we do the rest," with specs, drawings and addenda automatically sliced into conformed sections and revision-tracked sheets on upload, no template or reformatting step required before a team can start reviewing findings.

Project engineers, estimators and submittal reviewers already fluent in reading spec sections and drawing sheets generally find the findings format (a conflict or omission cited directly to its source page) maps closely onto how they already review documents manually, so the conceptual shift is mostly trusting the AI's first pass rather than learning a new document-review methodology from scratch.

Teams integrating Specset's output back into Procore or Autodesk Forma Build, or using its MCP/CLI interface for custom workflows, face a modest additional integration-configuration step, though the platform is built to sync both ways rather than requiring a full migration off an existing system of record.