The Wedge Into Offline Industries Is Fast, Cheap Data Collection

01 /The Concept

75% of construction projects run late. One key reason: nobody really knows what's happening on job sites.

Workers and foremen can't realistically write detailed progress reports at the end of every shift. Project teams, as a result, can't accurately track project status or manage it effectively.

The same problem applies to material deliveries – delays in supply are another major source of schedule slippage. And as a finishing blow, payment delays for materials and labor often aren't intentional but happen because of disorganized tracking of work completed and deliveries received.

Dexter set out to solve these problems by deploying specialized AI agents for real-time, detailed tracking of what's happening on construction sites.

The first AI agent acts as a field recorder. At the end of a shift, workers give it a verbal rundown of what they completed. The agent captures it, asks clarifying questions, and produces structured reports in the company's standard format – which it then routes to the general contractor and project team.

The other agents require integration with the company's existing project management and accounting systems.

Once connected, the second agent cross-references work reports against the project plan and the materials delivery schedule – marking completed plan items and updating the materials inventory. The general contractor and project team get a real-time picture of project status in their own systems, without any extra effort.

The third agent reconciles supplier invoices against what was actually delivered to the site – so contractors only pay for what arrived, and can flag suppliers immediately when deliveries fall short.

Although Dexter initially positioned itself as a construction management platform, the same AI agent family can manage other service-heavy operations, including equipment repair and maintenance – collecting voice reports from field technicians, converting them into structured records, and syncing with back-office systems.

Dexter went through Y Combinator last autumn, but only recently published its platform launch on the YC site.

The platform is already running in production at real companies.

02 /Why It Matters

Buildots ([covered here](/review/do-bolshih-deneg-ne-hvataet-odnoj-malenkoj-fishki)) is tackling the same construction management problem and has raised $166M total, with $45M closed in late May.

The basic model is similar to Dexter's: AI collects status information from the job site, compares it to the project plan, marks what's been completed, and flags risks of delays. The key difference is how the AI collects that information – not through voice check-ins, but through small cameras clipped to workers' hard hats. At the end of each shift, workers plug the camera into a laptop and the AI downloads and analyzes the footage.

What Dexter and Buildots share is the use of AI to fundamentally improve the data capture step – getting high-quality information out of the physical world so it can drive better decisions at the process level.

This same pattern shows up in several other startups that caught significant investor attention this year.

Siro ([covered here](/review/smotri-ka-ved-takie-prodazhi-tozhe-nuzhno-uluchshat)) raised $50M in May for an AI sales coach – specifically for field reps who sell installation, repair, and maintenance services in person. The rep records their customer conversations; the AI coach analyzes them and provides targeted feedback. Siro's data shows it increases closed deals by 36% and reduces rep turnover by 30% – because reps finally start earning properly.

XOi ([covered here](/review/tema-v-kotoroj-mozhno-i-horosho-zarabatyvat-i-horosho-prodatsja)) raised $230M in February for an app for equipment repair technicians. Point your phone at the equipment's nameplate, and the AI identifies the make and model, diagnoses likely failure modes based on described symptoms, and provides wiring diagrams and step-by-step repair guides.

Technicians currently spend an average of 2.5 hours a day searching for this information manually. XOi delivers a first answer in 4 minutes and resolves the task in 25. It also cuts callbacks and quality complaints by 40%.

03 /Opportunities

The physical world is where business actually happens. A huge share of all commerce is conducted offline, and running that commerce efficiently requires fast, simple, and cheap ways to capture what's happening on the ground. The quality and speed of that input data determines the quality of everything downstream – garbage in, garbage out.

The general direction: platforms for managing offline businesses whose core competitive advantage is an effective AI system for collecting field data.

The range of applications here is broad and often surprising. Earlier this year, Minimist ([covered here](/review/v-jetoj-teme-mozhno-sdelat-novyj-shopify)) launched an app for secondhand clothing stores: photograph an item someone brings in, and the AI identifies it, logs it in the inventory system, and suggests a price based on what similar items are currently selling for – all in 2 minutes.

What other businesses depend on fast, accurate collection of offline information? How can AI make that faster and more reliable? At what cost, and with what level of robustness?

That last point matters. XOi originally tried to solve the same problem using augmented reality glasses – and it didn't work. The hardware was too expensive and too fragile. Switching to phone cameras changed everything.