About 41% of the construction workforce retires by 2031.
The preconstruction side gets hit hardest, and mentorship saves only ~30% of what a senior estimator knows.
Codify the judgment before it walks out the door.
https://t.co/SrqGKX3wM1
Why a Mars-rover engineer turned down SaaS to build for the physical world.
@deepti_yenireddy on the Boon AI origin story with @reecejbarnes @adaptivebuilds on Builders, Budgets & Beers.
#Construction#AI
ICYMI: 41 seconds on why generic AI loses to vertical AI on a plan set.
Full episode of @deeptiyenireddy on @reece_barnes's Builders, Budgets & Beers: https://t.co/cJngAA4Noc
Industry-standard accuracy band on a conceptual construction estimate: -20% to +50%.
That's a $10M budget landing anywhere between $5M and $15M.
The fix isn't a new tool. It's making your firm's own historical project data accessible.
https://t.co/t2zFQLlPoz
Workflow. Bids feed contracts, contracts feed buyouts. An error on Tuesday surfaces on a jobsite six weeks later, after five other decisions trusted it.
Accuracy. One percent off on a mid-sized takeoff is a fifty thousand dollar swing in the bid. The wrong number looks identical to the right one on the page.
The CFO case for AI in preconstruction:
28% avg cost overrun.
~32% of it from estimating errors.
50 to 70% of estimator time on manual takeoff.
AI doesn't shrink the team. It lifts throughput 30-50%.
https://t.co/nRnaojstfs
"How many buildings are there in the US, how many plan sets? It's finite."
The hardest problem in construction AI is not the model — it's that the training data physically runs out.
Deepti on BBB Ep 77 with @verderamo.
37 narrow models, each stuck at its own ceiling.
1 foundation model, where every trade lifts every other trade.
The architectural choice that decides the next decade of construction AI.
Hyperscalers are spending half a trillion dollars on construction this year.
The labor isn't there to meet it without AI. From @deeptiyenireddy on BBB Ep 77.
https://t.co/cJngAA4Noc
"The only thing I can control is my pace of learning."
Deepti on the one thing every construction-AI roadmap gets wrong.
From Builders Budgets & Beers Ep 77 with @verderamo.
One moat used to be enough. Procore on networks. Veeva on data. Shopify on distribution.
Frontier models on a 6-month cadence changed the half-life of single moats. Three moats, not one.
"This year we all became managers, because we manage AI."
Deepti on Builders, Budgets & Beers Ep 77. The estimator workflow shift in one line.
https://t.co/cJngAA4Noc
Each era kept what the last one did well. Paper to digital removed the redrawing. Digital to AI removes the manual counting.
The rate-limiting step gets a layer of its own. The judgment stays with the estimator.