Construction

The number no AI drawing tool will publish

Construction AI markets generation speed. Nobody reports how much output survives engineer review unchanged, and that is the figure buyers need.

Five funded AI drawing products market generation speed. None publishes the number that decides their value: the share of generated output that survives professional review unchanged. Annija Eizenarma — a founder, investor and adviser to technology companies selling into legacy industries — describes engineers making "adjustments according to their preferences," and that phrasing is currently the only evidence anyone has.

The figure is knowable. These products could measure it. That none of them publishes it is the most interesting gap in the category.

Why does review survival matter more than generation speed?

Because it separates a drafting assistant from an error generator, and the two demo identically. If the engineer is applying preferences to broadly correct output, the tool is worth paying for. If the engineer is hunting for mistakes in output that looks correct, the tool costs more than it saves.

Annija Eizenarma's word choice is the tell. Preferences, not corrections. She was describing what she sees working in construction today:

"Even creating floor plans — now you can even prompt how you imagine your office to be, or the building to be, and it fully can be illustrated by AI. Like all of their electrical wiring, for example, all of the nuances can be illustrated and drawn with AI, where the engineer only needs to make some of the adjustments according to their preferences."

Electrical is the unforgiving case. A product built by someone who has never sat with a mechanical engineer will generate output that looks correct and fails review, and in this domain the buyer's professional liability attaches to the drawing.

What is the category selling instead?

Time. Augmenta generates complete electrical systems inside Revit and raised a $10M round in March 2025. Endra automates MEP engineering, fire alarm live with electrical, HVAC and plumbing indicated as coming, on a $50M Series A led by a16z in 2026. Higharc runs homebuilders from architecture through construction drawings on roughly $78M. Drafted returns floor plans and elevations in seconds on a $17.5M seed from Y Combinator Spring 2026. Maket generates residential plans from parameters without CAD skills.

Plans in seconds, systems designed automatically. Every pitch is a speed pitch, and speed is a cost-centre argument aimed at a professional-services buyer whose real constraint is the range of work they can win.

What would a better pitch sound like?

Capability gained rather than labour replaced. Annija Eizenarma's account of how this buyer is chosen is more useful than a feature list:

"How do people choose architects? They look at the awards, they look at their previous projects, they look at their expertise, for example. And if you can increase the expertise of an architect by, for example, adding more skills to his existing skill set, that's a big win for both sides."

Nothing in this product set is sold on letting a small firm credibly take on work that previously needed a specialist, or bid a project type it has not done before. That is a positioning error rather than a product error, which means it is fixable this quarter.

Why do founders miss this workflow in the first place?

Because they choose the market before they meet it. Asked why AI companies struggle in construction, Annija Eizenarma pointed at how the idea gets picked:

"Often I see startups coming up with business idea based on the market size, where they're sitting with AI and asking ChatGPT: 'Okay, ChatGPT, tell me how I can become a millionaire, which is the biggest market, where is biggest market share where it can build, where are many problems in which industry?' And I think it's a one-way, but not necessarily the smartest way."

Her prescription is one line: "They have to be really connected to the real world business which they are building for."

How long does this market take?

Longer than a seed runway assumes. Selling building automation in 2017, into a category she describes as not yet a real market, took her two years to reach a small salary, "because you have to educate the market first." And the education is not a deck:

"Often you do have to create a product to show them how it works before they can even believe it."

For AI drawing tools that cuts both ways. The demo is genuinely persuasive, and the distance between an impressive demo and output a licensed professional will sign is exactly where trust is won or lost.

Three changes worth making to the roadmap

Instrument review survival and publish it. The percentage of generated geometry the engineer of record signs without edits, by project type. The first company to report it honestly owns the comparison every buyer is trying to make privately.
Re-pitch on work won, not hours saved. Name the project type your customer could not previously bid, and price against that, not against drafting time.
State the boundary out loud. Where jurisdiction-specific code compliance lives, and what still requires a stamp. Candour reads as competence to a buyer carrying the liability.

If you cannot yet measure how much of your output survives review, that is the first thing to build. Your buyers are already measuring it, one project at a time, and they are not showing you the results.