Construction · Deep dive

Trunk Tools

AI agents for the built environment — ingest a project's drawings, specs, RFIs, submittals and schedules, then let field and office teams query them by text and hand tedious document work to agents.

emerging

The question that decides it: Trunk Tools' wedge is a construction-native intelligence layer (Cortex) that reads drawings, specs, RFIs and submittals across a project set and answers questions in the field by SMS. Does an independent document-AI layer stay defensible once Procore and Autodesk ship 'good enough' copilots on the data they already host — and do contractors trust agents on safety-critical documents when the company's own flagship case study logged 87% answer accuracy, not 99%?

My take ★★ Very interesting ★ Interesting − Maybe ✗ Not interesting

Snapshot

Trunk Tools sells an AI layer that sits on top of a construction project’s document chaos — drawings, specifications, RFIs, submittals, schedules, contracts, change orders, meeting minutes — and makes it queryable by the people who need answers fastest: the crews and superintendents in the field. Its best-known product, TrunkText, lets a worker text a project-specific phone number and get an answer with source citations in seconds; a newer intelligence layer, Cortex (launched June 2026), reads across drawing sets and powers a suite of workflow agents for drawing review, RFIs, bids and submittals. Founded in 2021 in New York by Dr. Sarah Buchner — a carpenter-turned-superintendent-turned-PhD — and Chris Boyd, it has raised roughly $70M, including a $40M Series B led by Insight Partners in July 2025, and says it is deployed on hundreds of active jobsites for GCs including Gilbane, Suffolk, DPR and HITT. Revenue reportedly grew 5x in six months off a small base. It is genuinely early, and the whole question is whether a standalone document-AI layer stays defensible once the platforms that already host the documents ship their own copilots.

Founding story

The founder is the pitch. Sarah Buchner grew up poor on an Austrian farm and started helping her father with carpentry around the age of 12. She did not observe construction from a business-school distance — she worked sites for more than fifteen years, rising from superintendent to project manager to group leader, and led BIM on what she describes as the first as-built BIM project in Europe, a roughly €300M job. Along the way she built a jobsite health-and-safety app, which pulled her toward applied construction technology, and she stacked graduate degrees onto a full-time building career: an MS in civil engineering and a PhD in data science / civil engineering aimed squarely at construction’s data problem, doing machine-learning document extraction before LLMs made it fashionable.

Her PhD is the origin of the company. Trying to apply data science to real projects, she kept hitting the same wall: construction data is unstructured, siloed across a dozen systems, and changing constantly — a drawing revision here, an RFI answer there, a spec buried in a 900-page PDF. She came to Stanford’s GSB in 2019, and in 2021 started Trunk Tools with co-founder Chris Boyd to make that document sprawl accessible and actionable. The differentiator she leans on is credibility: fifteen years on sites means she can tell, and claims her buyers can tell, the difference between construction-AI marketing fluff and something a superintendent will actually text on a Tuesday.

How it works

Follow one question on a jobsite. A superintendent standing at a wall needs to know the fire-rating detail for a specific partition, or the approved submittal for a beam, or whether a drawing was revised last week. Historically that means walking back to the trailer, opening the right PDF set, and hunting — 20 to 40 minutes of search-and-travel per question by Trunk Tools’ own accounting. Instead, they text a project-specific phone number. TrunkText returns an answer in seconds, with the source document attached so the worker can verify it.

Underneath is the harder part. Trunk Tools connects to wherever a project’s documents already live — Procore, Autodesk Construction Cloud/Forma, Box, SharePoint, Egnyte, Dropbox — and live-syncs the corpus. Cortex, the intelligence layer, reads and structures the whole set: it interprets drawing symbols, revision clouds and the relationships across hundreds or thousands of sheets, then links a drawing to the spec, RFI, submittal and change order that govern it. Because it uses LLM-style semantic understanding rather than keyword search, it answers a plain-language question even when the exact words never appear in the document. That connected foundation powers the agents — and it is safety-adjacent work, which is why the accuracy numbers below matter more than in most SaaS.

Product and business overview

The product is a platform with a field front-end and an expanding agent suite, all sitting on Cortex:

• TrunkText — the SMS assistant. Instant, cited answers from the project’s documents, accessible from any phone in the field. This is the wedge and the demo that lands.
• Cortex — the construction-specific intelligence layer (launched June 2026) that reads drawings and connects them to specs, RFIs, submittals, schedules and change orders. Trunk Tools frames it as “the brain,” the product of four years of development with large GCs, and says it powers seven agents.
• TrunkReview — scans drawing revisions and produces a visual overlay plus a written change list; claims a 20-sheet bulletin in about five minutes and 70-85% less manual review time.
• TrunkRFI — de-duplicates and drafts RFIs, killing ones already answered in prior responses or the document set, and updates status in the project management system.
• TrunkBid — turns a 4-8 hour trade-package review into a shorter decision by surfacing gaps, exclusions, alternates and scope silence per bidder.
• Plus submittal management and visual document browsing agents.

The buyer is a general contractor (and increasingly its trade partners) running large, document-heavy projects. Named customers include Gilbane, Suffolk, HITT, DPR, Harkins, Consigli, Torcon, McGough and Cleveland Construction.

Business model and pricing

Trunk Tools is enterprise SaaS sold by quote — it does not publish price points, and third-party catalogs (Capterra, Software Advice, GetApp, accessed 2026) all list pricing as custom/on-request. The commercial unit in practice is the project: deployments are described per-jobsite (“deployed on hundreds of projects”), and the field-enablement motion — getting crews to actually text the number — is treated as core to the model, with the Series B explicitly funding a field-enablement program. That points to a land-by-project, expand-by-account structure: win one job with one GC, prove the time savings, roll across the GC’s portfolio.

The honest read: with no published pricing and no disclosed revenue, the unit economics are a black box. The one external anchor is a Latka estimate of roughly $8.5M ARR for 2025 — an estimate, not a company figure, and Latka’s paired “$25.5M valuation” line is clearly stale (it is below cash raised), so treat both with caution. The reported 5x revenue growth in six months (company, July 2025) is impressive but is a rate off an undisclosed and probably small base. What a diligent investor cannot yet see: net revenue retention, per-project contract values, and whether portfolio-wide rollouts actually happen or deployments stay stranded on lighthouse jobs.

Traction over time

The most revealing traction is a single case study, because it contains a number the company did not have to publish. On Gilbane’s Baird Center project, Trunk Tools tracked roughly 21,000 documents; across 37 working days users asked 246 questions of TrunkText, and the project team validated 87% of the answers as correct. The company also cites 20-40 minutes saved per query, 30-plus minutes on many field questions (per a named superintendent), and $100,000+ of rework avoided per month. Gilbane went on to roll the agents across jobsites. The shape is a fast-compounding early-stage company with strong lighthouse references — and a candidly imperfect accuracy figure that frames the central risk.

Market analysis

The tailwind is real and large. Construction is chronically under-digitized — paper, PDFs, and siloed systems — while carrying an estimated $1T-plus in annual productivity losses (cited in Trunk Tools’ Series B materials, July 2025). The addressable software layer: the global construction software market was about $10.76B in 2025, and the narrower AI-in-construction market about $4.86B in 2025, projected to reach roughly $35.5B by 2034 at a ~24.8% CAGR (Fortune Business Insights, 2025). North America was ~39% of the AI-in-construction market in 2025.

The structural forces favor the thesis: a labor and expertise shortage, rising document complexity per project, and LLMs finally capable — in principle — of reading unstructured construction data. The counter-force is equally structural: the value is highest exactly where the documents sit, and those repositories are owned by Procore, Autodesk and Trimble, not Trunk Tools.

Competitive intel

The competitive set is crowded and, at the top, far better capitalized. Procore owns the GC software relationship and is extending Copilot across the modules it already hosts; the threat is bundling. Autodesk Construction Cloud/Forma owns the design-and-drawing backbone Cortex is trained to read, and is layering its own AI. Document Crunch was the closest pure-play on construction-document AI (contract/spec risk, 10,000+ projects, $37M+ raised) until Trimble acquired it in April 2026 — validating the category, removing a comp, and arming a third incumbent. OpenSpace and Buildots attack the jobsite from the reality-capture side (computer vision on site progress) rather than the document set, but compete for the same “AI on the jobsite” budget and each has raised well past $100M. Bild AI (YC W25, ~$3.5M) shows how cheap pointing vision+LLMs at blueprints is becoming. And the quiet default is in-house data teams and general-purpose LLMs.

Where Trunk Tools wins: it is construction-native and field-native — SMS access, cited answers, agents built around real GC workflows, and a founder who ran jobsites. Where it is exposed: its core asset is being attempted by the three platforms that already host those documents, plus a wave of cheaper startups. Its edge is a product-and-accuracy advantage, not yet a structural moat.

History and evolution

• 2021 — Founded in New York by Dr. Sarah Buchner and Chris Boyd, out of Buchner’s PhD work on extracting value from unstructured construction data.
• 2022-2023 — Early/seed capital; Innovation Endeavors and Fifth Wall back the company; TrunkText launches as the field-facing SMS assistant.
• Aug 20, 2024 — $20M Series A led by Redpoint Ventures (Innovation Endeavors following); total to-date ~$30M. TechCrunch profiles Buchner’s carpenter-to-founder arc.
• 2024-2025 — Named GC references accumulate (Gilbane, Suffolk, DPR, HITT, Torcon, McGough, Consigli); Gilbane rolls agents across jobsites; Baird Center case study logs 87% answer accuracy and $100k+/month rework avoided.
• Jul 24, 2025 — $40M Series B led by Insight Partners (Redpoint, Innovation Endeavors, StepStone, Liberty Mutual Strategic Ventures, Prudence); total ~$70M; revenue reported up 5x in six months.
• Jun 17, 2026 — Launches Cortex, the construction-specific intelligence layer, positioning drawings interpretation as the “hardest AI problem” and the platform’s brain; seven agents now sit on it.

No public crises or layoffs — the company is too young for a real history of stumbles, which is itself a caveat: scaling accuracy, support, and enterprise rollouts across a conservative industry is still ahead.

What people say

The case for. The field results are concrete and the references are blue-chip. On the Baird Center job, TrunkText fielded 246 questions in 37 days and the team validated 87% correct, with a named superintendent citing 30-plus minutes saved per question and the company claiming $100k+/month in avoided rework; Gilbane expanded from a pilot to jobsite-wide rollout. Trunk Tools’ framing that generic AI fails on construction drawings — and its willingness to publish a real accuracy number rather than a marketing 99% — reads as credibility to buyers. On Glassdoor (2026, small sample of ~9 reviews) employees describe a company that looks after people, offers growth, and has customers who genuinely love the product.

The complaints. The same sources carry the warnings. Glassdoor reviewers flag classic hyper-growth strain: growing pains from fast hiring, an intense deadline culture, and a push to “move faster” even when it risks lower-quality work. More importantly, the flagship accuracy stat cuts both ways — 87% correct means roughly one in eight answers was not, in a product whose answers touch load capacities, fire ratings and safety-critical specs, where a confident wrong answer can be expensive or dangerous. The structural critiques are sharper: pricing and revenue are undisclosed; the company is ~100 people and ~$70M raised against Procore, Autodesk and Trimble, all of which host the documents and can ship “good enough” copilots on data they already own; the moat is being commoditized by YC-stage startups; and adoption depends on getting a low-tech, skeptical field workforce to text a bot — which is why field enablement, not just model quality, is where the money goes.

Outlook: the open question

Trunk Tools works if a construction-native intelligence layer proves both more accurate and more trusted on safety-critical documents than the copilots Procore, Autodesk and Trimble bundle into the repositories they already own — and if that accuracy edge is durable rather than a temporary head start. It stalls if “good enough” answers, shipped for free inside software a GC already pays for, are good enough for the field, or if 87%-class accuracy caps how far crews will rely on it. That is the whole thing.

The bull case is unusually well-supported for an early company: a founder with fifteen years on jobsites and a relevant PhD, blue-chip GC references expanding from pilot to portfolio, a quantified pain (20-40 minutes and $100k/month of rework per project), and top-tier investors writing $70M of conviction. But the exposure is structural, not executional — Trunk Tools’ core asset (reading and connecting a project’s documents) is exactly what the three platforms that host those documents are now attempting, and the Document Crunch acquisition shows incumbents will buy rather than build. What would settle it in Trunk Tools’ favor: accuracy climbing convincingly toward the high-90s and staying there, net revenue retention above ~120% as agents stack and rollouts go portfolio-wide, and GCs still choosing it after Procore/Autodesk copilots ship. What would settle it against: field accuracy plateauing where trust breaks, deployments stranded on lighthouse jobs, and incumbents closing the gap for free. With ~$70M raised against far larger balance sheets, Trunk Tools does not have unlimited runway to prove the layer is a category and not a feature — the next raise, and its price, will tell you which one it is

How a challenger would attack it

Attack the accuracy gap, not the category. Trunk Tools’ most exploitable weakness is its own flagship number: 87% validated accuracy on the Baird Center job means roughly one wrong answer in eight, on questions about fire ratings and load capacities. A challenger would launch with a verification-first architecture — answers that ship with a confidence score, refuse to answer below a threshold, and route low-confidence queries to a human reviewer — and market the refusal rate as the feature. That reframes Trunk Tools’ candor as a liability. Second vector: the cost of entry has collapsed, as Bild AI’s ~$3.5M vision-plus-LLM blueprint reader shows, while Trunk Tools carries ~100 people, $70M raised, and a field-enablement program that is essentially paid human adoption labor. A lean challenger skips the SMS-training motion entirely and embeds inside the tools crews already open — Procore’s mobile app, plan-viewing software — rather than asking a skeptical workforce to text a new number. Third: undercut the opaque quote-based pricing with published per-project pricing that a mid-market GC can buy without a sales cycle, since Trunk Tools’ named logos (Gilbane, Suffolk, DPR) show it is fishing at the top of the market and leaving the long tail of regional GCs unserved.

Same playbook, new buyer

The playbook — live-sync a document corpus, structure it, answer field questions with citations — is not construction-specific; it is “regulated, document-heavy, safety-critical industry with a low-tech field workforce.” The most promising shift is down-market and sideways to trade subcontractors: Trunk Tools sells to GCs and only “increasingly” their trade partners, but a mechanical or electrical sub lives in the same spec-and-submittal sprawl with none of the enterprise sales attention, and per-project pricing sized for a $5M sub contract is a different product than a quote-based sale to Suffolk. Trunk Tools won’t follow easily because its cost structure — enterprise field-enablement teams, Insight-scale growth expectations — makes small-check, self-serve revenue unattractive. The second shift is geographic: Buchner’s own origin story is a €300M European BIM project, yet the named customers are all US GCs; European and Middle Eastern mega-projects have the same document chaos, different standards (Eurocodes, metric drawing conventions), and no incumbent copilot pressure from Procore, whose gravity is weakest outside North America.

Sources and further reading

• Trunk Tools Closes $40M Series B to Lead Construction’s AI Transformation (Insight Partners, July 2025)
• Trunk Tools Nets $40M in Funding Round Led by Insight Partners (Engineering News-Record, July 2025)
• Sarah Buchner started as a carpenter when she was 12 — now her AI construction startup has raised $20 million (TechCrunch, Aug 2024)
• Trunk Tools Raises $20M Series A Led By Redpoint Ventures (PR Newswire, Aug 2024)
• Trunk Tools Launches Cortex to Tackle Construction’s Hardest AI Problem: Drawings (Engineering News-Record, June 2026)
• How Gilbane used an AI tool to track 21,000 documents (Construction Dive, 2025)
• In the Field with Suffolk: How Trunk Tools Helps Crews Build Smarter (Trunk Tools, 2025)
• Trimble to Acquire Document Crunch (Construction Dive, April 2026)
• AI in Construction Market Size, Share & Industry Report (Fortune Business Insights, 2025)
• Trunk Tools Reviews (Glassdoor, accessed 2026)
• Trunk Tools Revenue 2025 estimate (Latka, 2025)