Construction AI BriefSubscribe →
Issue
№112
Pillar
Trend
Audience
GC ops
Dated
2026.07.25

Procore's AI agents just became a paid product. The feature worth noticing is the one built to outlast a retiring superintendent.

Procore moved its AI agents out of private beta into three general-availability packages on July 23 — and previewed a feature called Skills that lets a firm encode its own SOPs into the AI. That's a direct answer to the knowledge-loss problem every GC has when a veteran PM walks out the door.

ByConstruction AI BriefAbout this publication

Procore moved its AI agents out of private beta and into three paid, generally available packages on July 23 — and previewed a companion feature called Skills that lets a firm feed the AI its own standard operating procedures instead of relying on generic defaults. That's the more consequential half of the announcement. Pricing tiers are a normal product step. A feature explicitly built to capture "how the company operates" before the people who know how the company operates retire is a direct answer to a problem every GC already has.

What did Procore actually ship?

Three Digital Coworker packages, all now generally available: Starter Pack (five agents), Pro (the full library of 20 agents — covering things like Deep Search, Submittals, RFI, Daily Log, and Contract Review), and Enterprise, which adds Agent Studio for building custom agents on top of the pre-built set. Every tier includes Control Tower, an admin dashboard that reports AI usage and credit consumption by agent, project, or team member — the kind of visibility a controller needs before signing off on a company-wide AI line item.

PackageWhat's included
Starter Pack5 pre-built AI agents
ProFull library of 20 pre-built AI agents
EnterpriseEverything in Pro, plus Agent Studio for custom agents and deeper customization
All tiersControl Tower usage/credit tracking; Skills rolling out in August

What does Skills actually do?

Skills lets an organization upload its own SOPs, project standards, or plain-language process descriptions, and have Procore's agents apply that specific method — not a generic industry default — across every project. Procore SVP of AI and data Thiago da Costa framed the goal directly: AI output that "consistently reflects how the company operates," plus "a way to preserve institutional knowledge as experienced workers retire, change roles, or leave the organization." That's not a vague AI pitch. It names the exact failure mode a lot of GCs are already living through — a superintendent with 25 years of unwritten judgment calls walks out the door, and the next hire relearns all of it from scratch, on your schedule's dime.

What should a GC actually do with this?

Skills only works on what your firm has actually written down. If your submittal review process, your RFI escalation rules, or your daily-log standards live entirely in one PM's head, Skills has nothing to encode yet — and that's worth treating as the real prerequisite here, not the AI feature itself. Before August:

  1. Audit what's actually documented. Pull your current SOPs for submittals, RFIs, and daily logs. If they're thin or outdated, that's the gap Skills will expose, not fix.
  2. Ask what "generally available" costs. Procore hasn't published per-tier pricing publicly as of this announcement — get real numbers from your account team before assuming Starter Pack is the safe default.
  3. Keep the approval line where it is. When Procore first demoed these agents in June, it was explicit that the Submittal agent flags discrepancies and the RFI agent suggests edits — a human still decides. Nothing in this announcement suggests that changed, and Control Tower's usage tracking is about visibility, not an approval workflow. Don't let a Skills rollout quietly shift sign-off authority to the agent by default.

The Haskell case study Procore cited — submittal review time cut from seven days to 10 minutes — is a real number from a real customer, but it's one firm's result, not a benchmark. If you're evaluating Pro or Enterprise, ask your Procore rep for the same metric on a submittal volume close to yours before you budget the time savings into next year's staffing plan.

For more on what these agents do mechanically, see our earlier breakdown of Procore's AI agents from when they were still in private beta.

Construction AI Brief tracks what's actually shipping in construction AI tools, not just what's announced — new pieces most days at constructionaibrief.com.

FAQCommon questions
What is Procore Skills?
Skills is a Procore feature, previewed July 23 and rolling out across all Digital Coworker packages starting in August, that lets a construction firm teach Procore's AI agents its own standard operating procedures and project standards using plain-language prompts or uploaded documents — so the agents follow that firm's method instead of a generic default.
What are Procore's Digital Coworker packages?
Three tiers, all now generally available: Starter Pack includes five AI agents, Pro includes the full library of 20 agents, and Enterprise adds Agent Studio for building custom agents plus deeper customization. Every tier ships with Control Tower, which tracks AI usage and credit consumption by agent, project, or team member.
Do Procore's AI agents approve submittals or RFIs on their own?
No. When Procore first showed these agents in private beta in June 2026, it described them as flagging discrepancies and drafting summaries for a human reviewer to accept or reject — not approving or rejecting anything themselves. The July 23 announcement introduces pricing, packaging, and Skills, but doesn't describe a change to that approval structure.
Does Skills actually solve the problem of losing institutional knowledge when workers retire?
It targets that problem directly, per Procore's own framing — SVP of AI and data Thiago da Costa described the goal as AI output that reflects how a company operates and a way to preserve institutional knowledge as experienced workers retire or leave. Whether it holds up depends on how well a firm's SOPs are actually documented, since Skills can only encode what's written down or described, not what's only in a retiring PM's head.
What proof does Procore have that this works?
The company cited one customer example: Haskell reduced submittal review time from seven days to 10 minutes using the Submittals agent. That's a single vendor-supplied case study, not an independent benchmark, so treat it as a data point to verify against your own submittal volume rather than an expected result.
End of sheet — issue №112
Published · 2026.07.25
Project
Construction AI Brief
Dated
2026.09.07
Sheet
1 / 1
Rev
A
Published independently · constructionaibrief.com · © 2026Facebook·Privacy·About