Pirros has extended the AI agent it runs inside Revit from managing a firm’s content to reviewing a firm’s project. AEC Magazine reported on 7 October that the company has added an AI-powered Quality Management System that reviews an entire Revit project during design development against building codes, the firm’s own standards, project specifications and model performance, with findings assignable to a person for resolution — or to Mira, the company’s agent, to fix in the model. The launch framing and its timing are the magazine’s reporting, and this piece carries them as that; the capability page is live on Pirros’s own site, and most of the detail below is drawn from it.
A day later, on 8 October, the company published a second, separate update on its own blog: six new Mira skills, written by chief executive and co-founder Ari Baranian, aimed at documentation chores Revit makes a user perform one element at a time.
The two together describe a company moving in a specific direction, and the direction is more interesting than either announcement.
The Claim Is About the Reference, Not the Check
Pirros’s pitch starts from a premise about what is now table stakes: knowing the building code. What Baranian is selling is the other half of the comparison.
“Our QMS is different from most QA/QC drawing checkers in the market, which typically work with exported drawing sets that are disconnected from the Revit model data and often require users to input findings manually,” Baranian said, in remarks carried by AEC Magazine.
The company’s own product page states the argument more sharply in a subheading: “Every QA/QC checker knows the code. Only one knows your firm.” The mechanism behind that line is the company’s existing business. Pirros’s original product is a content platform — a firm’s Revit details, families, specifications and standards, ingested, searchable and version-controlled. The page’s claim is that this makes the review possible rather than merely better: “So when a detail doesn’t match the version your firm approved, we can show you, because we hold both the approved detail and the one in your set.”
It follows that there is no ruleset for a customer to author. “Nothing to upload, no ruleset to author,” the page says. “If your details and families live in Pirros, the system already has its reference.” Asked in its own FAQ whether a firm must build a ruleset first, the answer is: “No. If your details and families are in Pirros, the Quality Management System already has its reference. You tune weighting; you don’t author rules from scratch.”
What the system checks, per that page, runs across five categories: the firm’s vetted details and standards, against drift from the approved version; the set against itself, for “Callouts that point nowhere, keynotes that miss the legend, references that drift”; codes, listed as “IBC, CBC, and more. Whatever governs where you build, cited down to the section”; the project specification, once added; and the model’s own hygiene — “Naming, parameters, purge and audit state, file weight.”
It is not confined to Revit models. The FAQ states that “While Pirros is Revit-native, the Quality Management System can also read PDF sets. You can review a consultant’s drawings without their model ever leaving their office.”
One note for readers who go and look: the second illustrated review on that page carries the page’s own note, in its accessibility text, that “The drawing, the issues and the names are made up for this page.” Neither illustration is a case study.
Built for Someone Who Signs the Drawings
The control design is the part a practice’s technical lead will read first, and it is consistent across both announcements.
The product page lists four commitments: “Findings are surfaced, never auto-applied”; “Reviewers dismiss with a reason, and the system learns the firm’s tolerance”; “Full audit trail of what was flagged, who resolved it, and when”; and “Role-based permissions on who can change a firm standard”. The blog post describing the six new Mira skills sets out the same discipline in four rules — Mira shows what it found before touching anything, nothing is written until the user confirms, the whole batch undoes with a single command, and it “checks every change afterward and names anything it skipped, with the reason.”
The post is also specific about refusals, which is unusual enough in product copy to be worth quoting: “It won’t invent a style, create a parameter, ungroup elements or guess at an ambiguous match. Those go back to you.”
That is a product shaped by the liability its buyer carries. A quality manager cannot stamp a drawing on an agent’s confidence score, and both documents are built around keeping the decision with a person.
The Six Skills, and Why They Are Called Skills
The 8 October skills are small by design. Each takes a job Revit forces a user to repeat element by element and does it across a sheet, a view or a model in one pass: renumbering details on a sheet, auditing a view’s annotation against a template, bulk text-note edits, bulk parameter setting, centring room tags, and flipping grid heads.
The specifics show domain knowledge rather than a generic wrapper. On renumbering, the post notes that “Details marked SIM or N.T.S. are flagged, not overwritten.” On annotation audits, that Mira “matches styles by their properties rather than their names” and “reports a style your project doesn’t have instead of inventing it.” On parameters, that it “warns when a parameter sits on the type and would change every instance” and “fills parameters that already exist; it won’t create new ones.” On room tags, that a room’s tag and its location point are separate objects in Revit, so centring one by hand leaves the other behind and the next view’s tag lands off-centre again.
Baranian’s stated reason for packaging these as fixed skills rather than leaving them to prompting is the most quotable idea in either announcement, and it is an argument about firms rather than about models:
“Asking a general AI to do the same job twice likely results in two different approaches, especially when different team members make the same request.”
“A skill fixes the method: the same checks, the same questions and the same report, whoever runs it.”
Alongside the skills, the post says Mira now plans multi-step jobs, reads PDFs as well as Revit models, and summarises older parts of long conversations so chat length is unbounded. It also states that “Mira is free to download for Revit 2025 and later”, and that this version “builds on the earlier version we released in May 2026” — the company’s news page dates the original Mira launch to 15 May 2026.
The Round That Named This
The useful context is ten months old and comes from the company. When Pirros announced its Series A, it said what the money was for.
The 15 December 2025 release records a $15 million Series A led by Elephant Ventures, announced from Los Angeles, following a $2 million seed in 2023 with participation including Y Combinator, FundersClub and Twenty Two Ventures, plus angels including former Autodesk chief executive Carl Bass, Joseph Walla of HelloSign and Ryan Sutton-Gee of PlanGrid. The Series A added ServiceTitan co-founders Ara Mahdessian and Vahe Kuzoyan as angels, with Y Combinator and FundersClub returning. The company states in that release that it “has been profitable since its seed round” — its figure, reported as such.
And then this sentence: “This Series A funding is an investment that Pirros will channel directly into continued innovation for architects and engineers, including quality management.” The QMS is not a pivot. It is the thing the round said it would build, shipped roughly ten months later. The founding team is described in the same release as Ari Baranian and Peter Johann, structural engineers turned entrepreneurs.
What the Customer Evidence Does and Doesn’t Show
Pirros names 32 practices in the logo row on its homepage, among them Arup, WSP, Thornton Tomasetti, Populous, Michael Baker, Clayco, Clark Nexsen, DCI Engineers, KPFF and Lake|Flato Architects. That roster is for the platform as a whole, not for the new QMS.
For the QMS specifically, one firm is named. Trent Piazzoni, an associate and project architect at Number TEN Architectural Group in Winnipeg, is quoted in AEC Magazine’s report: “When you’re working on a school, experience tells you many of the items that require particular attention. This [QMS] platform can reinforce that process by consistently highlighting relevant items that may warrant another look.” The bracket is AEC Magazine’s — Pirros’s own version of the same quote reads “This platform”.
The same firm is the subject of a customer case study on Pirros’s own site, which is where most of what is publicly known about the QMS in use comes from, and which should be read as the vendor’s account of its own customer rather than as independent verification. It says Number TEN “joined the beta program to test Pirros’s latest launch” earlier this year, and adds a second voice from the firm: Jessica Cembroski, an associate and project delivery manager, who runs the setup — linking projects through Autodesk Construction Cloud, scheduling the reviews and choosing which codes each review checks against. For Number TEN that means “the Alberta Building Code and the National Building Code, rather than the American rulesets most tools ship with”, which is a more substantial detail than it looks: the QMS page’s own list of codes reads “IBC, CBC, and more”, both American.
The case study also carries the only specific result published anywhere this publication could find: “Running the set against the current building code surfaced sloped glazing and safety glass requirements that had been omitted from the set” on a project close to going out to bid. One finding on one project, reported by the vendor, is not a measurement of the system — but it is at least a concrete claim rather than a percentage.
Number TEN is a real Canadian practice with offices in Winnipeg and Victoria and a portfolio that includes schools, consistent with Piazzoni’s remark, and it is not among the investors Pirros names in its seed or Series A announcements. The relationship is the ordinary one of a vendor and a beta customer, disclosed here because the quotes in both the trade report and the case study come from it.
Read the quotes closely and they are careful ones. The phrases “may warrant another look” and “can reinforce that process” describe a second pair of eyes, not a verdict engine — and Cembroski puts it in exactly those terms in the case study: “It’d be another set of eyes for us.” That is the same claim the product pages make, and a narrower claim than a reader skimming “AI-powered QA/QC” would assume.
On commercials, the product page is partly open. Checking the health of a firm’s own content “comes with the platform”, while “Reviewing an active project set before it goes out is an add-on that runs on top of it. Seats are unlimited either way; the project review is metered on the sheets you review.” Prices are not published: “Pricing depends on your firm and how much you’re checking. Talk to us.” The page also states SOC 2 Type II certification, GDPR and CCPA compliance, and “Zero AI training on your data.”
Analysis: Two Routes Into the Same Room
The following is this publication’s reading.
The day before the Pirros report, this publication covered Structured AI shipping a platform that loads IFC or native Revit alongside the PDF set so that a finding on a sheet resolves against the element that produced it. The two companies have arrived at the same destination — review that runs on the model rather than on a flattened export — from opposite ends of the problem.
Structured AI started as a drawing reviewer and moved inward to the model, with the check library as the asset; it raised a $4.2 million seed in June. Pirros started by holding the firm’s content and moved outward to the set, with the reference library as the asset. Both are in Revit. Both refuse to auto-apply. Both stake their differentiation on context the other has to acquire.
Which asset compounds faster is a genuine open question, and the answer is probably not the same for every buyer. A check library is portable across firms and gets better for everyone each time it grows. A reference library is per-firm and gets better only for that firm — but it is also far harder for a competitor to take away, because taking it away means persuading a practice to re-ingest its whole accumulated detail library somewhere else. That is switching cost dressed as a feature, and it is one of the oldest reliable moats in professional software.
The risk in Pirros’s position is the same one the whole cohort carries, and it is worth stating plainly rather than at the end of a sentence. Building inside Revit means building inside Autodesk’s application, on Autodesk’s API, in a year when Autodesk is shipping its own construction agents and when a rival BIM vendor is opening its own authoring tool to agents over MCP. Depth in a firm’s accumulated content is a real defence against a generic checker. It is a thinner one against whoever ends up owning the model.
What no outside reader can assess from either announcement is the thing that decides whether any of it works: whether the review catches what matters without burying a team in findings it has to dismiss. The dismiss-with-a-reason loop suggests the company knows that is the risk. Neither announcement measures it.
