Every Y Combinator batch seems to produce another AI takeoff startup. In December 2025, Bobyard raised a $35m Series A led by 8VC for a tool veteran construction operators had barely heard of. The reflex from the industry is to call this dumb money. But there is a coherent VC logic that says funding the 50th company in a category is not just defensible, it is the right move. Foundamental’s Patric Hellermann laid it out on the latest Bricks, Bucks & Bytes. Here is why VCs keep writing the checks, why operator expertise can blind you to the bet, and what the PlanGrid precedent says about how these waves usually end.
If you watch the construction technology landscape long enough, you start to notice a pattern. Every six months, another AI takeoff tool gets announced. Another seed round. Another Series A. Another founder explaining that they have rebuilt computer vision for blueprints and that their model is 99% accurate and that estimators using it can submit three to five times more bids.
Veteran operators look at this and lose their minds. Dustin DeVan, who built BuildingConnected and sold it to Autodesk for $275m, vented about exactly this on the latest Bricks, Bucks & Bytes. He knows the CEO of every meaningful takeoff company in the market. He had never heard a single customer mention Bobyard, despite its $35m Series A from 8VC in December 2025.
And yet the money keeps flowing. Foundamental’s Patric Hellermann, the third voice on that episode, offered a defence of the VC behaviour that is worth taking seriously, even (especially) if you think the operator critique is right. The two views are not actually in conflict. They are looking at different parts of the same elephant.
If you are an early-stage VC indexing the AI rewrite of an entire industry, the question is not “is this product good.” It is “is this team going to discover what the product should be faster than anyone else.” Those are very different bets.
The operator critique is mostly correct
Takeoff is not on the critical path
Start with what the operators are getting right, because the steelman only works if you take their objection seriously.
Quantity takeoff is the process of pulling measurements off a set of drawings to feed into a cost estimate. It is genuinely tedious work. A senior estimator can spend 100+ hours tracing footings and grade beams on a concrete bid. AI computer vision can compress that to minutes. None of that is in dispute.
The problem, as DeVan and Hellermann both argued, is that takeoff is not where estimates go wrong. The big misses in construction pricing are on unit costs (what something will actually cost to build, given current labor and materials) and on missed scope items (things that should have been priced but weren’t). Quantities are an emergent property. As Hellermann put it, you will be 20% wrong too high on some items and 20% wrong too low on others, and the portfolio largely averages out. The dollar gap between an owner’s expectation and a GC’s bid is not driven by whether someone counted footings accurately. It is driven by everything around the count.
This is the real reason takeoff feels like a strange place to keep concentrating venture capital. Faster takeoff makes the existing process less painful for estimators. It does not move the critical path of pre-construction. Anyone who has lived inside an estimating department knows this in their bones.
Why VCs fund it anyway
The “directionally right + best team” thesis
Now the steelman. Hellermann’s framing is one of the cleaner articulations of early-stage VC strategy you will hear on a construction tech podcast. It works like this.
As an early-stage investor, you cannot know which products will win. The market is too noisy, the customer behavior too unpredictable, the technology too volatile. What you can do is be directionally right about the opportunity, then back the team you think will out-execute everyone else in figuring out what should actually be built.
AI has supercharged that logic. Anything a company builds today will likely be rewritten in six to twelve months. The codebase is a depreciating asset. The product is a depreciating asset. The only thing that compounds is the team’s velocity at discovery: how fast they ship, how fast they learn from customers, how fast they kill bad bets and double down on good ones. If you believe takeoff is directionally part of the AI rewrite of pre-construction, and you find a team that will out-discover everyone else in that space, you take the bet. You are not buying the product. You are buying the option on whatever that team eventually builds.
You want to be in the founders that can build an organization that is the fastest at discovery, creating optionality, and seizing it.Patric Hellermann, General Partner, Foundamental
This explains a behavior that looks irrational from the operator’s chair. If you have a fund large enough to write 50 checks across a category at 0.5% allocation each, you do not need every bet to work. You need one to blow up. Indexing the AI-takeoff category at the founder level is a coherent strategy, even if any individual company in your basket has a weak business case.
The PlanGrid precedent
Crowded categories can still produce a $875m exit
The historical question Martin asked on the episode is the right one: when a category gets this crowded this fast, does it ever produce an outlier winner? DeVan’s answer was instructive. When the iPad arrived around 2011-2012, there were 50+ startups building mobile plans for the jobsite. The category looked saturated. Most died. Procore got huge. PlanGrid sold to Autodesk for $875m in 2018. Fieldwire did well. The crowd did not prevent the winners.
The drone wave of 2012-2017 went the other way. Vast amounts of venture money flowed into construction drone startups. Other than DJI (which is not really a construction company), nobody built a generational business. The category got commoditized before anyone built a moat.
So which way does AI takeoff break? Honestly, nobody knows. But that uncertainty is precisely the point. A VC backing Bobyard or any of the YC takeoff cohort (Rudus, Fresco, and others) is not claiming to know. They are claiming that the team they backed is the one that will figure it out. If they are wrong about the team, the bet dies. If they are right, the crowded category is irrelevant.
Get the next deep dive in your inbox
News, deals, and unfiltered takes on construction tech. 300+ issues and counting.
Join 2500+ ReadersThe expert-blindness problem
Knowing the market can hurt you at the earliest stages
Hellermann made one more point that founders should sit with. The danger of being an expert in a category is that your expertise gives you very fast, often correct, reasons to dismiss new entrants. The unit economics don’t work. The customers aren’t switching. The differentiation isn’t there. The KPIs they are pitching (“we help you bid faster”) are not the KPIs the market actually rewards.
All of that can be true and still miss the bet. If the team executes well enough, they will discover something the expert did not predict. The expert is grading the company against today’s product. The investor is grading the team against next year’s product, which doesn’t exist yet.
This is also why a lot of construction veterans struggle as early-stage VCs in their own industry. The mental habit of pattern-matching against what has not worked is exactly the wrong instinct when the underlying technology stack is rebuilding itself every two quarters. Hellermann was open that even he has to consciously override his own expertise on some bets. His example: backing Bedrock Robotics, where he was not convinced the market was big enough but trusted the team enough to write the check anyway.
Where the steelman runs out
Late-stage logic looks different
The “directionally right + best team” thesis is an early-stage argument. It does not extend cleanly to Series B and beyond.
At those stages, the question shifts from “can this team discover what to build” to “can this company defend a position in a real market.” That is where the operator critique starts to bite harder. If the 50th takeoff company has Series B money but no differentiation, no enterprise references, and a product that’s substitutable with three competitors, the math gets ugly fast. As KP Reddy of Shadow Ventures has argued, the Series A bar in ConTech has moved to roughly $5m ARR. By the time you are raising real growth capital, you need real revenue against a real moat. Founder velocity alone does not get you there.
This is also why the venture industry’s response to “category looks crowded” is asymmetric by stage. At seed, crowding is fine. At growth, crowding is fatal. The takeoff category is currently sitting in the awkward middle, where the seed and Series A rounds are still flowing freely but the eventual sorting has not happened.
Operator view versus VC view
| Dimension | Operator view (DeVan) | Early-stage VC view (Hellermann) |
|---|---|---|
| Primary question | Is the product solving a real critical-path problem? | Is this team the one that will out-discover the category? |
| Reaction to crowding | Red flag, no differentiation | Indexing signal, back the founder |
| Time horizon | Today’s customer behavior | What the team will build in 12-24 months |
| How to evaluate the bet | Customer references, unit economics, ARR | Founder velocity, market direction, optionality |
| Where the view holds best | Series B onwards, real market sorting | Pre-seed and seed, before the product exists |
| Failure mode | Late to changes invisible from inside the industry | Spray-and-pray with capital, no underwriting discipline |
Frequently asked questions
Bobyard raised a $35m Series A in December 2025, led by 8VC, with participation from Pear VC, Primary Venture Partners, Tishman Speyer, RXR, Caffeinated Capital, and Merrick Ventures. The company started in landscaping takeoff and is expanding into drywall, electrical, HVAC, plumbing, and framing. (Source)
Because takeoff isn’t where preconstruction goes wrong. The biggest cost misses come from unit price errors and missed scope items, not from inaccurate quantities. Faster takeoff compresses estimator hours, which is a real efficiency gain, but it doesn’t change the critical path of pre-con timing or the gap between owner expectations and GC bids. (Source)
It can be, and Hellermann was clear that the framework requires careful management of internal bias. The thesis only works if you have genuine conviction in the team and you are sized to absorb the failures that come with indexing. A small fund spraying capital across a crowded category without that discipline is not running the same playbook, even if the surface behavior looks similar.
Around 2011-2012, dozens of startups raced into mobile plans for the jobsite when the iPad arrived. The category looked saturated. Most died. PlanGrid sold to Autodesk for $875m in December 2018. Procore became one of the most valuable construction software companies in history. A crowded category did not prevent a generational outcome. The drone wave that followed went the opposite way and produced no comparable winner. Outcomes are not predictable from crowding alone. (Source)
The exact number per batch varies, but the trend is clear. AI-native construction tools appear in nearly every recent cohort. Public examples include Rudus (concrete takeoff, claims 70% time reduction) and Fresco (Division 8 doors, frames and hardware takeoff). Broader context: nearly half of YC’s Spring 2025 batch were classified as AI agent companies. (Source)
If you are reading this and treating it as a green light, you have the wrong takeaway. The point of the steelman is that VCs are funding teams, not products. If you don’t have founder-market fit, unusual velocity, and a clear-eyed view of why takeoff alone is not enough, you are entering a category where capital is no longer the constraint and differentiation is. The Series A bar in ConTech has moved to roughly $5m ARR. (Source)
Most likely: aggressive consolidation. Either established estimating platforms (Ediphi, Togal, Stack, Toggle) absorb the better takeoff startups as features, or the AI takeoff layer becomes a commoditized API that everyone integrates with. A small number of teams will discover an adjacent wedge (cost intelligence, scope detection, design clash analysis) that turns the takeoff product into a foothold for something bigger. The rest will quietly disappear when the next funding round doesn’t close.
Related Articles
How BuildingConnected was sold for $275m
A venture capitalist’s hot takes on construction tech
Business Wire: Bobyard raises $35m Series A led by 8VC
Autodesk: Completes PlanGrid acquisition for $875m
Y Combinator: Construction startups directory
PitchBook: YC going all-in on AI agents
ENR: Ediphi and the case for cloud-based precon estimates
Crunchbase News: Hellermann on polar-bear investing
