Why Banks Are Beating Builders to Agentic AI (And What That Says About Construction’s Data Problem)
Banks are deploying agentic AI in production faster than general contractors. That sounds like a glitch in the matrix. Banks are heavily regulated, slow-moving, risk-averse. GCs are not. Yet Built Technologies launched its first generally available agentic product on the lender side of its business in February 2026 because that was where the data was usable. The reason is regulation: it forced banks to document their workflows for thirty years. Construction did not. The result is a documentation debt that has now become an AI debt, and it is about to define which contractors win the next decade and which ones fall behind. This is your wake-up call.
The construction industry has a story it likes to tell about itself. It is slow to adopt technology, but resourceful. It runs on relationships and gut judgment, not paperwork. The work happens in the field, and the field does not care about your documentation standard. There is a kind of pride wrapped up in that, and for a long time it was an acceptable trade-off. Software was a back-office concern.
That trade-off just expired. The same lack of documentation that made construction culturally distinctive is now the single biggest reason builders cannot deploy AI agents the way other industries are starting to. And the contrast is sharpest with the industry construction tends to look down on: financial services. Banks are slow, banks are bureaucratic, banks make construction look fast. But banks are also winning the agentic AI race, and they are winning it for a reason that should make every GC executive uncomfortable.
The reason is that AI agents do not run on data alone; they run on documented decision logic. They need to know not just what happened, but what an experienced operator would do next and why. Banks have spent thirty years writing that down because regulators forced them to. Construction spent the same thirty years leaving it in the heads of senior PMs. The bill is now due.
The agentic AI gap between banks and builders is not a technology gap. It is a documentation gap. And documentation is downstream of regulation. Construction’s cultural preference for tacit knowledge has just become a structural disadvantage at the worst possible moment.
The Built Technologies tell
Why a fintech-leaning construction company launched its agent on the bank side first
If you want to see this asymmetry in production, look at how Built Technologies sequenced its agentic rollout. Built serves three customer segments: lenders, owner-developers, and general contractors. The company has roughly $300 billion of active projects on its platform. When it shipped its first generally available agentic product in February 2026, the agent went to lenders. Not GCs. Not developers. Lenders.
Built’s CEO Chase Gilbert explained the reasoning on a recent episode of our show, and it is worth quoting directly. Banks, he said, are heavily regulated, and because they are regulated their policies and procedures are obsessively documented. They are held accountable to following those policies by their regulators. AI responds well to well-documented process: you can hand an agent a written workflow and it will execute it, fall back appropriately when something is off, and even resolve its own problems when the documentation is rich enough. Owner-developers and general contractors, by contrast, run on a mix of documented process and what Gilbert called “extremely well-documented to barely documented at all.” On the GC side specifically, Built ships not full agents but “micro automations”: small bounded automations that wrap together to lighten the workflow rather than replace it. The product is shaped by what the data allows.
This is the most honest signal in the market right now. A company that builds for the entire construction stack has decided, based on actual deployment data, that the same underlying AI works very differently depending on whose documentation it is reading.
The better you can begin to document what you want to happen, the agents just respond so much better. In the absence of that documentation, it’s kind of hard.Chase Gilbert, CEO, Built Technologies
What an agent actually needs from your business
Why “we have data” is not the same as “we are agent-ready”
There is a common misconception that AI agents need data. They do, but data alone is not the bar. What they actually need is the decision logic that connects the data to action. Many business workflows depend on tacit knowledge, the kind of “I just know” judgment experienced operators apply without being able to articulate it. When you try to encode that workflow into an agent, the missing rules and instructions become startlingly obvious.
Consider what a typical mid-cycle decision on a construction project looks like. A subcontractor’s invoice arrives with a slightly different scope description than the contract line item. The contract administrator looks at it, knows from past dealings with this sub that the description usually maps to the same line item, knows that the field super likes this sub and will be unhappy if payment is delayed, and approves it. No rule was followed. The decision was good. It is also completely opaque to any system that was not the person making it.
Now multiply that across procurement, RFI handling, change orders, safety incident routing, lien waiver collection, schedule rebaselines, and a hundred other workflows. Construction runs on thousands of these implicit calls per project. The bank, meanwhile, has a 47-page procedure document for how to approve a draw request, who reviews it, what triggers escalation, and what the default action is when documentation is incomplete. The bank can hand that to an agent. The contractor cannot.
How construction got into this hole
Three structural causes, none of them malicious
Nobody decided that construction should not document its workflows. The lack of documentation is the cumulative outcome of how the industry actually operates, and the causes are worth naming because they tell you why fixing it is not just a software problem.
- Project-based work fragments knowledge. Every project has a different owner, designer, subcontractor mix, and site conditions. A “process” that worked on one job is partially obsolete on the next. The pressure to document is low because the documentation does not transfer cleanly. Knowledge stays with the person who lived through it.
- Tight margins punish overhead. Average GC margins sit at 2-3%. Process documentation feels like back-office expense in an industry where back-office expense is what kills you. The same firm that would never spend a week documenting its draw process will spend a week tracing a misallocated joint check, because the second one is reactive and the first one is “optional.”
- No regulator forced the issue. Banks document because OCC and FDIC examiners show up and ask. Pharma documents because the FDA does. Construction’s regulators care about safety outcomes and code compliance, not about whether your change order workflow has a written SOP. So the workflow stays in the senior PM’s head, and the senior PM is now retiring.
The retirement wave makes this worse
Tacit knowledge does not transfer when the holder leaves the room
If construction had unlimited time to fix its documentation debt, this would be a long-term concern, not a wake-up call. It does not have unlimited time. 41% of the current construction workforce is projected to retire by 2031, per the National Center for Construction Education and Research. The senior PMs and supers who hold the tacit knowledge are walking out the door in numbers the industry has never seen.
That retirement wave is bad on its own merits. Bridgit’s 2026 Construction Workforce Benchmark Report shows the average project team is now 36.4% rookies, climbing to 56% on teams of 51 or more. The institutional knowledge is leaving, and the people inheriting it are by definition new to the company. They will not pick up the tacit playbook just by being on site. Most of them will not get the chance to.
The grim version of this is straightforward: the senior people whose decision logic could have been documented and fed to agents are the same senior people who are about to retire without documenting it. Every retirement is a deletion. Construction teams already lose 18% of project time just searching for information that exists somewhere in the company, but cannot be found when needed. The next ten years are going to compound that.
Download our research library on AEC tech and the data shaping it
Reports, case studies, and operator playbooks for builders, founders, and investors. Free to download.
Browse the ReportsThe asymmetry, side by side
Why agents fly in one industry and crawl in another
| Dimension | Banks | General contractors |
|---|---|---|
| Regulatory pressure to document | OCC, FDIC, internal audit | Code compliance only |
| Workflow knowledge format | Written SOPs, decision trees | Tacit, in senior PMs’ heads |
| Process consistency across teams | High, enforced by audit | Low, project-by-project |
| Data structure | Standardized, structured fields | PDFs, emails, spreadsheets |
| Tenure of decision-makers | Stable, codified roles | 3.7-year median for senior PMs/supers |
| Agentic AI readiness | Mature, in production | Early, “micro automations” only |
What to do about it
Documentation is no longer back-office hygiene. It is the AI strategy.
The good news is that the fix is unglamorous and well within reach. Construction does not need to become a bank to deploy useful agents. It needs to externalize the decision logic of its experienced operators while those operators are still in the building. The firms that do this in the next 18 months will pull away from the ones that do not.
The investment angle
Where the smart money is positioning
For investors watching ConTech, the documentation gap creates a clean thesis. The most defensible AI products in construction over the next five years will be the ones that own the documentation layer first. Construction productivity is roughly where it sat in 1950 while every other industry has tripled, and a meaningful share of that gap is captured by exactly this issue: process knowledge that cannot scale because it cannot be written down.
That is a market opportunity hiding in plain sight. The companies that figure out how to capture senior operators’ decision logic at the speed of conversation, codify it into a usable schema, and feed it to agents will be sitting on the only data moat that actually matters in this industry. Built has a head start on the lender side. There is no equivalent leader yet on the GC side. There will be.
For incumbents, the implication is harder. The Procores, Autodesks, and Trimbles of the world have the customer base and the data scale, but they also inherited the same documentation problem their customers have. The vendor that ships the best documentation-and-agents product, regardless of whether they started in construction or arrived from another industry, will define the category. Trade contractors are already moving faster than the GCs above them, partly because their workflows are tighter and easier to document. That pattern is going to repeat at every level of the stack.
The bigger picture
This is the industry’s productivity problem, finally crystallized
Construction has been told for years that its productivity problem is about technology adoption. Buy more software, train more people, embrace digital workflows. That framing was always a little off. The deeper issue was that the industry’s knowledge infrastructure was unwritten, and unwritten knowledge does not scale. Software did not fix it because software inherited the same gap.
Agentic AI exposes the gap in a way nothing else has. When the bottleneck is the documentation, not the model, the cure is structural, not technological. The contractors who internalize that and start writing things down today will look back in five years and recognize this as the inflection point where they made the right call. The ones who keep telling themselves construction is too unique to document will look back in five years and wonder why their competitors got faster while they did not.
Banks are not winning at agentic AI because they are smarter. They are winning because their regulator made them write everything down. Construction’s regulator is now retirement, and it is just as effective at forcing the issue, but only if you act before the knowledge walks out the door.
Frequently asked questions
Banks are required by regulators to maintain heavily documented policies and procedures. That documentation is the substrate AI agents need to operate effectively. Construction firms have no equivalent regulatory pressure on internal workflows, so process knowledge tends to live as tacit knowledge inside experienced operators rather than as written SOPs. The asymmetry is not about technology access; it is about documentation maturity. (Source)
Tacit knowledge is the implicit, experience-based judgment that operators apply without being able to fully articulate it. A senior PM knows how to handle a scope-ambiguous invoice not because of a written rule, but because of years of seeing similar cases. AI agents cannot replicate that judgment because the rules were never written down. Externalizing tacit knowledge into a documented format is the prerequisite to deploying agents on any non-trivial workflow. (Source)
Built’s lender customers have decade-long histories of documented policies and procedures driven by regulatory compliance. That documentation made it possible to train an agent on real workflows and ship it in production in February 2026. On the GC and developer side, where workflow documentation is much less mature, Built ships “micro automations” instead of full agents: small bounded automations that wrap together rather than a single agent doing an entire job. The product strategy is shaped by what each customer’s data allows. (Source)
41% of the construction workforce is projected to retire by 2031. The senior PMs and supers leaving are the same people who hold the tacit knowledge that could be documented and fed to agents. Each retirement effectively deletes a portion of the institutional decision logic. The window to externalize that knowledge while those operators are still in the building is closing rapidly, which is why documentation has become a workforce-strategy issue rather than a back-office hygiene issue. (Source)
A micro automation is a small, bounded automation that handles a narrow part of a workflow with clear expected outcomes. Built uses the term to describe how it ships AI to general contractors and developers, where full agentic workflows are not feasible because the underlying process is not documented. The idea is to wrap many small automations together to make the work feel lighter, rather than asking an agent to take over a complete job function on undocumented process. It is the right model for any contractor whose workflows are still partly tacit. (Source)
Start with the workflows where decision frequency and dollar impact are both high: draw cycles, change orders, RFI routing, lien waivers, safety triage. Use your most experienced operators as the source of the decision logic, framed as architects rather than subjects of replacement. Deploy any agent in audit mode first, so it observes and flags rather than acts, and let the flags surface where your documented process diverges from the actual one. Treat the documentation effort as a retention bridge rather than a tech project. (Source)
The tacit knowledge problem exists in every industry deploying agents, which is why 85% of enterprise AI pilot failures trace back to data quality issues. But it bites construction harder for two reasons: project-based work fragments process documentation more than continuous operations do, and the retirement wave is concentrating the loss of institutional knowledge into a shorter window. The principles are universal; the urgency is specific to construction. (Source)
Related Articles
78% of Trade Contractors Use AI: Adoption Is Way Ahead of What You Think
“We Need a Moonshot”: Why AI in Construction Is Focused on the Wrong Problem
Construction’s Productivity Problem: 75 Years of Data That Should Bother You
The Insurance Game Is Changing and You Should Be Aware
Unite.AI: Why AI Agents Run Into a Knowledge Problem, Not a Technology One
VentureBeat: Three Disciplines Separating AI Agent Demos From Real-World Deployment
Bridgit: AI Construction Statistics for 2026
Bridgit: Rookie Ratio and How to Balance Experience on Construction Project Teams
Inteq Group: Managing Knowledge Worker Resistance When Deploying AI Agents
