For ten years, construction software has made the same promise: write your report on a tablet during the meeting or the walkthrough, and leave with a finished document. No more evenings catching up. No more trips back to the office. The promise is real. We just think it answers the wrong question.

Every project is one of a kind: its contract, its site, its crews, its schedule. We made that case in The Problem Wasn't There: the further you get from the site, the more structure you add to make up for the distance, and forms designed at the office end up as PDFs nobody reads. The real question isn't how to write faster. A meeting report has one job: get every trade to act on the priorities, in service of the building. That means staying focused on the site while you write it, and making it readable for the people who have to act.

Three generations of the report

The word processor: free, but stuck at a desk. Word and Excel are a wonderful place to work… at the office. On site, they have nothing to offer. You take photos on your phone, and getting them into the document is such a pain that you give up. As for the PDF, in tiny type, it's hard to read on a phone: pinch to zoom, then scroll sideways.

Tablet apps brought photos, drawings, and auto-generated reports. Their promise: finish the report in the meeting. That comes at a price:

• A device on its own, offline. Tablets are expensive and rarely have a data plan. Hence the focus on offline mode: you work alone, cut off from the rest of the team.

• One more thing to carry. It takes two hands, and on site you need a strap just to get around.

• A screen that steals your attention. For every sentence, you find the right field, scroll, dictate or type, fix it, while the meeting and the work move on.

• A form that boxes you in. Every entry takes several taps, and even if you dictate, you lose the freedom to write in the margins the way you would in Word.

You leave with a finished report, but you were less there. The tablet digitized the form; it didn't change who adapts to whom. The superintendent fills in the vendor's fields, in the vendor's order.

The third generation, the one there is building, flips that around. You talk the way you do in the meeting, and the software puts the information where it belongs. AI prepares, people approve, the report reads on a phone with no account to create, and you know who read it.

Our customers back this up. Word and Excel come up in every one of our case studies, and some teams also relied on a specialized app, one of them for fifteen years. One firm had ruled an app out because it didn't want to pay per-user licenses. Corentin Le Gars of Cosécante puts it from the reader's side: "When you get a report in Excel, you think: this company hasn't moved up to the next level."

Be There: capture without thinking about it

For notes, our principle fits in two words: Be There. Be present. On a walkthrough, to see what needs seeing. In a meeting, to hear what's playing out between the trades. Every time you look down at a screen to fill in a field, you lose some of that attention.

So we ask for nothing at capture time. One long recording, phone in your pocket or on the table. Photos with captions. In a basement or a dead zone, the note syncs when the signal comes back. Offline isn't a way of working anymore. It's a technical detail.

The best voice-note tools figured this out for video calls: no bot in the meeting, full attention on the people. We take it where they don't go: the site meeting around a table, and the walkthrough, in the mud or the basement, with photos.

Then AI adds structure, using the project's context: previous reports, company names, trade packages, acronyms. It sorts action items into the report's sections, and a person approves. As we wrote in When AI Frees People, the real cost isn't time. It's the energy it takes to get started.

Writing during the meeting is still an option, and some of our customers do it. At Chabanne, 80% of the report gets written in the meeting. But not alone, head down over a tablet: on a projected screen everyone can see, while the voice note captures the rest. You no longer have to choose between listening and writing.

Control isn't progress

Form apps and construction ERPs are systems of record. Their job is to fill a structured database so someone can check the work is being tracked. On mobile, forms pile up, harder to find and longer to fill out. On the web, modules pile up to cover every process in the business, heavier and heavier to administer. Then the database churns out rigid PDF extracts for every recipient.

Saving time is what every piece of software promises. The question is: for whom? Time saved by the people doing the checking gets paid for twice. In attention, by the people entering data on site. And in rigid, hard-to-read PDFs, by the trades who have to act. That's the wrong fight. And since every tool wants to be the center of the world, these databases don't talk to each other well: data piles up without feeding any improvement loop. More checks don't move a project forward any more than more security cameras do.

We think these systems will become back offices, filled in automatically from the real work by new generations of software and AI. None of them will dominate: the more specialized and open they are, the more useful they'll be.

Tools, not gadgets

Word and Excel have lasted for decades for a simple reason: above all, people want a simple, open-ended tool to work together and move the project forward. A form app, by contrast, ends up as a Swiss Army knife with a hundred blades. Each one is handy, but the handle gets too thick to hold, and you can't find the right blade.

As long as it sticks to a few forms, it works. At Finalcad, which we co-founded, dropping a pin on a drawing to close out a punch list item was ruthlessly effective. Every module we added after that (quality, safety, environment, scheduling, messaging) cost us some of that edge. Every feature has to earn its place. Products rarely fail for lack of features.

But the deeper problem is the approach itself. A form tries to fit the work into checkboxes. A project is one of a kind, and chopping it into boxes distorts it. A form is right for one specific task, not for reporting on a whole project. Word is a good knife: it only makes documents, but it makes all of them. there keeps that freedom and adds what Word can't do: construction blocks that understand tasks, attendance, progress, and headcount, and an AI that knows the project. No user permissions to manage. No licenses to count.

An easy slope, all the way to your own workshop

Construction knows the pride of building: driving past a building and saying, "I built that." It's the same drive behind the excitement about AI. People love to make things. A form never gives you that pride: you look down and submit to a check. A software company should aim for the opposite. Every user proud of what they share, and becoming a creator at their own pace.

Advocates of malleable software call it a gentle slope from user to creator: every step takes a little effort, never a cliff. That's how we designed the document. Import your Word or PDF report. Add a tasks block, then an attendance block. Bring in your branding in one click. Save a template, reusable snippets, instructions for the AI. In the end, everyone has their own workshop, and a report that reflects their project and their company. One they're proud to send.

At CICAD, an ISO 9001-certified subsidiary of Ingerop, the report hinged on five milestone dates specific to the company. A form would have flattened them. We kept them. "Adapting the tool to our requests made all the difference," says Philippe Maubert, project director. Since October 2025, every new CICAD project starts on there.do.

A solid core and shell, a custom fit-out

This instinct is part of a broader shift. At Notion, it's become a principle: software should bend, not the user. OpenAI just made it the heart of ChatGPT: its model composes the interface for every answer. The interface adapts to each person while the structure stays put. In construction terms: core and shell first, then fit-out.

Not everything should be fluid, because a report is binding. An interface that reshuffles every time you read it is hard to verify, and a contractual document has to read exactly the same three years from now. On a project, some things have to hold no matter what: the record of what was said, sent, and read; the locked document and its versions; proof of distribution; and data hosted in Europe, never used to train a model. That's our core and shell. The fit-out adapts to each customer: branding, templates, blocks, notes, instructions for the AI. Malleable where you capture and write. Fixed where you sign.

The PDF and our interfaces even share an ancestor: PostScript, invented by two Xerox PARC alumni to describe a page, in which Steve Jobs saw the language of a screen. Display and print are two sides of the same problem.

So one report gets three renderings:

• a Letter-size PDF: formal, traceable, printable proof;

• a mobile e-report: readable with no account, no zooming, no sideways scrolling, with photos that open in one tap;

• the same document on desktop: interactive and comfortable to work in.

The PDF is core and shell. The screen is fit-out.

AI that suggests, and decides nothing

AI is only as good as the context you give it. On a project, that context already exists: it's the meeting report, week after week, shared across trades. A form chops it into fields. A document keeps it whole, and that's what the enhanced note and AI·there read before suggesting anything.

Regenerate a report in one go and it changes with every version. You can't tell what moved. That's the flaw in most AI tools: you can't change one thing without everything else changing too. So AI·there and note merging don't rewrite the report. They suggest changes one at a time, and you accept or reject each one. Only what changed on site changes in the text.

Your notes belong to you. They're private by default, and the audio and raw transcript never leave the person who recorded them. Only the text you've approved goes into the shared report.

A meeting report is a contractual document. A misattributed trade or a wrong date can be expensive. So there's AI decides nothing and never sends anything. What's new shows up in blue from one week to the next, the document locks when it's sent, and the proof PDF records what was sent, to whom, and who read it. We don't claim AI never makes mistakes. We make sure a mistake is visible before it goes out.

A Renaissance, not an assembly line

At OpenAI DevDay 2026, Sam Altman contrasted two visions of AI: an industrial revolution that turns people into cogs in a machine, or a Renaissance that gives them more power. Construction naturally picks the second. Every building is one of a kind, and the value lies in the judgment of the people running it.

Dario Amodei, co-founder of Anthropic, puts it another way: however capable AI gets, it runs into the speed of the physical world. A project moves at the pace of concrete, trades, and weather, not paperwork. AI doesn't speed it up. It removes the friction around it.

AI doesn't replace expertise. Even the day it can write everything, the architect is still the one who signs and takes responsibility. AI takes away the tedious parts (copying, updating, formatting, chasing people down) and gives back what the tablet took: the time and attention to be on site.

The tablet tried to save time by making you type faster. We'd rather you didn't have to type at all. You talk, you take photos, you stay present. The report follows, and the people who get it know, right on their phone, what they need to do. Saving time isn't the goal. The goal is for every trade to act on what matters, in service of the building.

We tell the story of what this approach changed for our customers in one year, with the numbers, in There Is a Loop. And if you want to see your own report in there, send it to us: we'll send back a version ready to go. www.there.do/switch