AI tools that do one job are a dead end.
Menu
Which one are you?Your card (0)Plain-text version ↓Sources and credits

Scene unavailable. The poster frame and the words in HTML

White paper workflow cards tumble through the dusk over a street of roofs; one card has a gold edge

The claim

AI tools that do one job are a dead end.

We call them point solutions. In roofing, they are a dead end for the roofers who buy them, the companies that build them and the investors who fund them.

The lessons of "Vertical AI is dead" that held up, and a test you can run on any tool

Here is what held up from the video that said so, and a test you can run on any tool.

Our answer is no, for three different people, for three different reasons.

What this shows

Labels on this scene

  • Picture Opening film: generated illustration · mood only.

Is a one-task AI tool a good business, a good buy or a good bet?

Our answer is no, for three different people, for three different reasons.

You know the claim the page has to back up.

An AI point solution is a tool that does one task: answer the phone, write an insurance supplement, quote a roof.

This page is the short, moving version of our research paper. Every claim on it comes from the paper, which lists each source and its limits.

Foundation Projects builds a platform for roofing companies, so we are not neutral.

  • Our reading AI point solutions for roofers are a dead end: for the software companies that build them, the roofers who buy them and the investors who fund them. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Foundation position Foundation Projects is building a founder-aligned platform for roofing companies, so it is not neutral on this question. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).

Read the research paper →

Where the claim came from.Follow the next part ↓

Scene unavailable. The poster frame and the words in HTML

A roofing company owner in a truck cab at dawn, looking at a phone

We checked a video: “Vertical AI is dead.”

  1. We checked a video that said “Vertical AI is DEAD.” Devin Kearns posted it in September 2026: 15 minutes, about AI tools built for one industry.

  2. We turned it into 24 claims. Then we checked each one against primary sources.

  3. 4 held. 12 held in part. 8 could not be sourced. None was contradicted.

  4. This page uses only what held, in full or in part, in its corrected form.

We turned the video into 24 claims and checked each against primary sources: 4 held, 12 held in part, 8 could not be sourced, none was contradicted.

What this shows

Labels on this scene

  • Picture Truck film: generated illustration · mood only.

Does the argument hold up?

We turned the video into 24 claims and checked each against primary sources: 4 held, 12 held in part, 8 could not be sourced, none was contradicted.

You know this page uses only what held.

We cite the video as the source of the framing, not as evidence.

A separate fact-check moved four of our grades down before the paper was published.

  • Attribution The argument began as a 15-minute video essay by Devin Kearns, "Vertical AI is DEAD", published 30 September 2026. The paper cites it as framing, not as evidence. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading Foundation turned the video into 24 checkable claims and graded each against primary sources: 4 held as stated, 12 held in part, 8 could not be sourced and none was contradicted. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Project decision The paper uses supported claims, and partial claims only in their corrected form. Unsupported claims are not used. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).

Watch the video we checked → Download the claim ledger (CSV) →

The playbook it describes.Follow the next part ↓

Scene unavailable. The poster frame and the words in HTML

A grid of twenty paper workflow cards; one lifts into a beam of light, its left edge gold

The playbook and what absorbed it

The playbook: find one expensive task. Add AI.

  1. The playbook: find one expensive task and add AI. For three years it was a common AI startup pitch: pick one costly task, add AI, sell the savings.

  2. In roofing, the tools pitched to us most are an AI phone receptionist, an insurance supplement writer and an instant-quote tool.

  3. They are real tasks with real money behind them. Some of them work.

  4. A narrow tool is an honest way into an industry. The question is what happens next.

A narrow tool built around one costly task. In roofing: a phone receptionist, a supplement writer, an instant quote.

What this shows

Labels on this scene

  • Picture Wedge film: generated illustration · mood only.

What is an AI point solution?

A narrow tool built around one costly task. In roofing: a phone receptionist, a supplement writer, an instant quote.

You can name the three roofing wedges.

In 2023 Sequoia described the first wave of AI apps as copilots with “undifferentiated pitches” and “terrible” user retention.

  • Our reading For three years a common AI startup pitch ran on one move: find one expensive, well-defined workflow, add AI, sell the return. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading While models could not do much alone, a narrow tool with a clear payback was an honest way into an industry. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading The roofing wedges pitched to Foundation most often are an AI voice receptionist, an insurance supplement writer and an instant-quote tool. They are real tasks with real money behind them, and some of them work. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Attribution In September 2023 Sequoia described the first wave of AI apps as copilots with undifferentiated pitches and terrible user retention. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
What happened to the thing they were built on.Follow the next part ↓

Scene unavailable. The poster frame and the words in HTML

Paper workflow cards slide along a dark table into a soft white light

ChatGPT and Claude keep taking in these tasks.

  1. ChatGPT and Claude keep taking in what these tools sell. The dates are public.

  2. 2023: ChatGPT searches your PDFs, and anyone can build a custom assistant without code.

  3. 2024: Anthropic launches an open standard to connect assistants to business systems.

  4. 2025: OpenAI's voice model can answer phone calls. Claude gets skills for spreadsheets, slides and documents.

  5. By October 2026: business tools plug straight into ChatGPT.

  6. Tools do not vanish. But what made them special is now easy to get.

File search, connectors, skills, phone calls and business plugins all moved inside the general-purpose assistants between 2023 and 2026.

What this shows

Labels on this scene

  • Picture Light film: generated illustration · mood only.

What happens to a wedge when the thing it was built on stops being scarce?

File search, connectors, skills, phone calls and business plugins all moved inside the general-purpose assistants between 2023 and 2026.

You see why a wedge loses its edge.

The dates come from the companies' own pages: OpenAI (custom GPTs, 2023; phone calling for its voice model, August 2025; plugins page, by October 2026) and Anthropic (the Model Context Protocol, an open standard for connecting assistants to business systems, 2024; Agent Skills, 2025).

  • Official description November 2023: OpenAI's GPTs let anyone build custom assistants without code, and ChatGPT searches attached PDFs. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Official description November 2024: Anthropic launched the Model Context Protocol, one open protocol to connect assistants to business systems. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Official description In August 2025 OpenAI's realtime voice model added phone calling over SIP. In September 2026 its newest voice model went generally available at $0.05 a minute for the voice session, with model and tool usage charged separately. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Official description October 2025: Anthropic shipped Agent Skills for Excel, PowerPoint, Word and fillable PDFs. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Official description By October 2026 OpenAI's plugins page invited business users to connect their apps "right into ChatGPT", from finance and sales to customer support and small business. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading The timeline does not say every tool disappears. It says the thing a tool was first valued for stops being scarce. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
Which one are you?Follow the next part ↓

The same problem ends in a different dead end for each of you.

Which one are you?

What this shows

Labels on this scene

  • Picture Films: generated illustration · generated illustration.
Roofing company owners

The owner's path

What does a shelf of tools do to my company?

Scene unavailable. The poster frame and the words in HTML

A roofing office manager on the phone at a desk with three screens and sticky notes

Your dead end, and the test

One job. Nine steps. A tool for each.

  1. One roof job has nine steps. Follow it from the first call to the last review.

  2. Lead, inspection, estimate, claim, contract, production, materials, invoice, review: nine steps.

  3. In the point-solution pitch, each step gets its own tool. Each tool keeps its own copy of the homeowner.

Each tool keeps its own copy of the homeowner, and the owner becomes the person who wires them together.

What this shows

Labels on this scene

  • Picture Office film: generated illustration · mood only.

What does a shelf of tools do to a roofing company?

Each tool keeps its own copy of the homeowner, and the owner becomes the person who wires them together.

You see the integration work a shelf of tools hands back to you.

The nine steps are illustrative, not a count of what any roofer uses.

  • Our reading One roofing job passes through nine steps: lead, inspection, estimate, insurance claim, contract, production, materials, invoice and review. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading Follow one job from the first call to the last review. In the point-solution pitch each step gets its own tool, and each tool keeps its own copy of the homeowner. An owner who buys a tool per step becomes the integrator. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
Nine tools means you wire them together.Follow the next part ↓

Scene unavailable. The poster frame and the words in HTML

Nine paper cards pinned on a corkboard in a roofing office at night, tied together with tangled string

Nine tools means you wire them together.

  1. The gaps between tools are handoffs. Nine tools, nine copies of the same homeowner.

  2. Buy a tool for every step, and you become the person who wires them together.

The gaps between tools are handoffs, and the owner who buys a tool per step becomes the integrator.

What this shows

Labels on this scene

  • Picture Corkboard film: generated illustration · mood only.

Who connects nine tools?

The gaps between tools are handoffs, and the owner who buys a tool per step becomes the integrator.

You see the integration work a shelf of tools hands back to you.

The nine steps and tools are illustrative, not a count of what any roofer uses.

  • Our reading In the point-solution pitch each step gets its own tool, each tool keeps its own copy of the customer, and the gaps between tools are handoffs. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading Follow one job from the first call to the last review. In the point-solution pitch each step gets its own tool, and each tool keeps its own copy of the homeowner. An owner who buys a tool per step becomes the integrator. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
Who answers when a tool is wrong.Follow the next part ↓

Scene unavailable. The poster frame and the words in HTML

A desk phone lights up in an empty roofing office at night; rain on the window

If the tool is wrong, someone still answers.

  1. If the tool is wrong, someone still answers. An AI receptionist that answers a call can record it.

  2. Florida needs every party's consent to record a call. Texas needs one party's.

  3. Claim rules stay with you too. Texas and Florida decide who may act for a policyholder, and neither statute mentions software.

  4. In our reading, the duty to tell callers they are talking to AI lands on you, whatever tool answers. OpenAI tells developers to make it clear, unless it is already obvious.

The duties land on whoever runs the phone line or works the claim, whatever tool does the task.

What this shows

Labels on this scene

  • Picture Office film: generated illustration · mood only.

Who carries the risk of a tool's mistake?

The duties land on whoever runs the phone line or works the claim, whatever tool does the task.

You know the tool does not carry your duties.

Texas bars a contractor from acting as a public adjuster on its own jobs; Florida bars a licensed contractor from public adjusting unless it is also licensed as one.

Statute text read in October 2026. This is not legal advice.

  • Official description Florida requires every party's consent to record a call; Texas requires one party's. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Official description Texas bars a contractor from acting as a public adjuster on its own jobs, and Florida bars a licensed contractor from performing public adjuster services unless it is also licensed as one. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading In Foundation's reading, the duty to tell callers they are talking to AI lands on whoever runs the phone line, whatever tool answers it. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading Texas and Florida decide who may speak for a policyholder on a claim. Neither statute mentions software. Rules like these protect the licensed person, not the point solution that drafts the letter. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
What held up from the video.Follow the next part ↓
Software builders

The builder's path

Does my product have a moat?

Scene unavailable. The poster frame and the words in HTML

A thin layer of paper cards pressed between two huge dark stone slabs, light leaking through the gap

Roofing AI tools get squeezed from both sides.

  1. A roofing AI tool gets squeezed from both sides. Above it, the assistants ship the skill: reading files, taking calls, connecting systems.

  2. Below it, the data owners keep their data. Verisk's license says: “We own all Price Data provided with the Services.”

  3. EagleView says it owns its aerial imagery. In April 2026 it announced its own AI product.

  4. In between sits the tool: a workflow and a screen. In our view the gap is packaging, and packaging is the easiest part to copy.

Between the assistants above, which ship the skill, and the data owners below, which keep their data and ship their own AI.

What this shows

Labels on this scene

  • Picture Slab film: generated illustration · mood only.

Where does a roofing point solution sit?

Between the assistants above, which ship the skill, and the data owners below, which keep their data and ship their own AI.

You see the squeeze on a wedge from both sides.

Among the vendor pages we opened, none shows OpenAI, Anthropic or Google selling a roofing receptionist, supplement writer or quote tool. That absence is the gap point solutions fill.

  • Our reading A roofing point solution sits between two owners: the assistants above ship the capability, the data owners below keep their data and ship their own AI. In Foundation's view the gap is packaging, and packaging is the easiest part to copy. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Official description Verisk's license says "We own all Price Data provided with the Services", and in May 2026 it announced connectors that bring its analytics into Claude. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Official description EagleView states that "All aerial imagery is collected and owned by Eagleview", and in April 2026 it announced its own agentic AI product. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
What a vertical label is worth.Follow the next part ↓

Scene unavailable. The poster frame and the words in HTML

An empty glass meeting room at dusk; the wall screen is switched off

The industry label is only where it started.

  1. The industry label is only where a company started. Harvey started with law firms; it now reports more than 1,000 customers in 60 countries.

  2. In 2025 its share of revenue from corporate customers went from about 4% to 33%, its chief executive told TechCrunch.

  3. Our first-hand account: in October 2026 Harvey demoed to us, a roofing platform, not a law firm. The offer: about $20,000 a year, no usage charges.

  4. That is not Harvey becoming a law firm. It is a product sold as vertical, offered to an outsider on terms that look like a software license.

Harvey, the leading legal AI company, offered its product to a roofing platform on terms that look like a software license.

What this shows

Labels on this scene

  • Picture Meeting room film: generated illustration · mood only. The picture is a generated illustration, not Harvey's product or the demo.

Does a vertical label protect a product?

Harvey, the leading legal AI company, offered its product to a roofing platform on terms that look like a software license.

You see that a vertical label does not keep a product in its vertical.

The demo is our first-hand account of one conversation. We have not confirmed it with Harvey, there is no public price list to check it against, and terms vary by buyer.

Harvey's March 2026 post also says “Nothing about our commitment to law firms is changing.”

  • Vendor-reported Harvey's founders made "the explicit decision to focus our attention on law firms". Harvey now reports more than 1,000 customers in 60 countries. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Attribution Harvey's chief executive told TechCrunch that about 4% of revenue came from corporates at the start of 2025 and 33% did by November. Its March 2026 post also says "Nothing about our commitment to law firms is changing." Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Vendor-reported In October 2026 Harvey demoed its product to Foundation, a roofing platform, not a law firm, and offered it for about $20,000 a year with no usage charges. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading This is not Harvey becoming a law firm. A vertical label describes where a vendor started. It does not describe where the vendor will sell. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
What a vendor needs after the introduction.Follow the next part ↓

Scene unavailable. The poster frame and the words in HTML

A paper workflow card on dark wet sand at dusk as a small wave washes over it

After the first sale, you need a moat.

  1. After the first sale, you need a moat. In 2023 Sequoia warned that data moats are “on shaky ground” and saw workflows and user networks lasting.

  2. After the introduction, a vendor needs one of three things: a place in the workflow that is expensive to leave, a network, or an outcome it stands behind.

  3. A tool with none of them is the first thing the next assistant release makes optional.

  4. A system the owner runs the whole business on is different. It is not a one-job tool; in our view it is a candidate foundation.

A place in the workflow that is expensive to leave, a network, or an outcome the vendor stands behind.

What this shows

Labels on this scene

  • Picture Tide film: generated illustration · mood only.

What lasts after the introduction?

A place in the workflow that is expensive to leave, a network, or an outcome the vendor stands behind.

You can check your own product against the three things that last.

  • Attribution Sequoia warned in 2023 that data moats are "on shaky ground" and saw workflows and user networks building more durable advantage. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading A wedge buys an introduction. After it, a vendor needs a place in the customer's workflow that is expensive to leave, a network around it, or an outcome it stands behind. A tool that holds none of them is the first thing the next assistant release makes optional. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading A roofing system where the owner actually runs the business, the CRM and field system of record, is not a point solution. It is a candidate foundation. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
What held up from the video.Follow the next part ↓
Investors

The investor's path

Is a roofing AI wedge a good bet?

Scene unavailable. The poster frame and the words in HTML

A stack of a hundred sheets of heavy paper with one hair-thin sheet of gold leaf catching the light

Each tool gets about $1,200 a year.

  1. The average U.S. roofing company with employees took in about $2.9 million in 2022.

  2. It spent 0.22% of that on software. Counting computer services, 0.33%.

  3. Payroll was about 87 times the software line.

  4. Software and computer services: 0.33%, about $9,570 a year. Split eight ways (our assumption), one tool earns about $1,200 a year from a company.

About a third of one percent of a roofing company's revenue goes to software and computer services. Payroll is about 87 times the software line.

What this shows

Labels on this scene

  • Picture Paper stack film: generated illustration · mood only.

How big is the budget a roofing tool sells into?

About a third of one percent of a roofing company's revenue goes to software and computer services. Payroll is about 87 times the software line.

You know the ceiling on a seat in roofing.

U.S. Census Bureau, Economic Census 2022, roofing contractors (NAICS 238160). The figures describe 2022, before most roofing AI tools existed, and count expensed software only.

Eight tools is an assumption. We found no measure of how many tools a roofing company runs.

  • Derived arithmetic The average U.S. roofing firm with employees took in about $2.9 million in 2022. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Official estimate In 2022 U.S. roofing contractors with employees reported $69.9 billion in receipts. They paid $13.5 billion in payroll and expensed $154 million of software: 0.22% of receipts, or 0.33% counting purchased computer services. Payroll was about 87 times the software line. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Derived arithmetic At the measured share, spread across an assumed eight tools, the average roofing company spends about $1,200 a year on any one of them ($1,196). A vendor selling one tool at that price needs 8,360 customers to reach $10 million a year, 35% of all roofing firms with employees, and 83,595 to reach $100 million, 3.5 times the number of roofing firms with employees. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18); seat-arithmetic.
What that means for one tool.Follow the next part ↓

35of 100

would have to buy one tool for its maker to earn $10 million a year

What this shows

Exact value: 34.7 per 100 (of every 100 roofing companies with employees would have to buy one tool for its maker to earn $10 million a year).

Each figure stands for 1 in 100 U.S. roofing companies with employees (24,082 in 2022). Not real people.

Each figure stands for 1 in 100 roofing companies with employees. Not real people.

Gold figures: the share a one-tool vendor would need as customers to earn $10 million a year, at about $1,200 a year each.

The $1,200 assumes the measured software share spread across eight tools. Eight is an assumption.

  • Roofing companies with employees
  • Would have to buy one tool for $10 million a year

Labels on this scene

  • Arithmetic on an assumed eight tools, not a forecast The number on screen.

How many customers does a one-tool vendor need?

You see why every wedge is pitched as the first step into something wider.

Roofing companies with employees, out of every 100: each figure is 1 of 100 roofing companies
ShotRoofing companies with employeesNot used in this sceneNot used in this sceneWould have to buy one tool for $10 million a yearExact value
of every 100 roofing companies with employees would have to buy one tool for its maker to earn $10 million a year65 of 1000 of 1000 of 10035 of 10034.7 per 100

Arithmetic on an assumed eight tools, not a forecast

  • Derived arithmetic At the measured share, spread across an assumed eight tools, the average roofing company spends about $1,200 a year on any one of them ($1,196). A vendor selling one tool at that price needs 8,360 customers to reach $10 million a year, 35% of all roofing firms with employees, and 83,595 to reach $100 million, 3.5 times the number of roofing firms with employees. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18); seat-arithmetic.
  • Official estimate About 24,000 roofing firms in the U.S. have employees (24,082 in the 2022 Economic Census). Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading That is why every one of these wedges is pitched as the first step into something wider. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Official estimate In 2022 U.S. roofing contractors with employees reported $69.9 billion in receipts. They paid $13.5 billion in payroll and expensed $154 million of software: 0.22% of receipts, or 0.33% counting purchased computer services. Payroll was about 87 times the software line. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).

At about $1,200 a year, $10 million takes 35 of every 100 roofing companies.

  1. There are about 24,000 roofing companies with employees in the U.S. Each figure here is 1 in 100.

    35 of 100 would have to buy one tool for its maker to earn $10 million a year

  2. At about $1,200 a year each, $10 million takes 8,360 customers: 35 of every 100.

    35 of 100 would have to buy one tool for its maker to earn $10 million a year

  3. $100 million would take 3.5 times the number of roofing companies with employees.

    35 of 100 would have to buy one tool for its maker to earn $10 million a year

  4. That is why each of these wedges is pitched as the first step into something wider.

    35 of 100 would have to buy one tool for its maker to earn $10 million a year

Where the money that agrees with this is going.Follow the next part ↓

Scene unavailable. The poster frame and the words in HTML

An aerial view at dawn of a service company's yard with rows of white vans

Some big investors are buying service companies instead.

  1. Some big investors are buying service companies instead of tools. General Catalyst raised $1.5 billion in 2024 to own and run them, PitchBook reported.

  2. Long Lake, an AI holding company it backs, closed its $6.3 billion purchase of American Express Global Business Travel on 29 September 2026.

  3. None of that proves the strategy works. Closing a purchase is not showing the model pays.

  4. The question is not whether the tools work. It is who keeps what they create.

To owning and running service businesses, not to funding another seat. None of that proves the strategy works.

What this shows

Labels on this scene

  • Picture Yard film: generated illustration · mood only. The picture is a generated illustration, not any company named here.

Where is capital going instead?

To owning and running service businesses, not to funding another seat. None of that proves the strategy works.

You know the question to ask the next pitch.

Long Lake is backed by investors including General Catalyst.

This is research, not investment advice, and not an offer of securities.

  • Attribution General Catalyst writes that technology builders capture more value by owning and operating the service businesses; it raised $1.5 billion in 2024 for that strategy, according to PitchBook. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Partner announcement Long Lake, an AI holding company backed by General Catalyst among others, completed its $6.3 billion purchase of American Express Global Business Travel on 29 September 2026. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading None of that proves the strategy works. Closing a purchase is not the same as showing the model pays. What the deals show is where the capital is pointed: at owning the service, not at selling it a seat. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading The question for an investor in roofing AI is not whether the tools work. It is who keeps what they create. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading Graded supported: becoming the service captures the budget that software never touches. General Catalyst states it as its thesis and is funding it. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
What held up from the video.Follow the next part ↓

Scene unavailable. The poster frame and the words in HTML

Graded index cards in a grid on a desk at night under one lamp

The video's big point held. Some details did not.

  1. The video's big point held. Some details did not. Here are nine of its 24 claims, with how each one held.

  2. Held: the big assistants already do much of what one-job tools sold. File search, skills and plugins shipped inside them, and a shared standard was launched to replace one-off connectors.

  3. Held: owning the service captures the budget software never touches. General Catalyst states it as its thesis and is funding it.

  4. In part: buyers want one foundation, not a shelf of tools. We found no buyer survey. Our reading: each tool keeps its own copy of the customer.

  5. Too strong: the claim that each era of AI tools makes the last one obsolete. Plugins are a ChatGPT feature again, and workflow automation is still sold.

  6. Too high: the claim that software is about 2% of a law firm's revenue. Clio measures under 2% of expenses, not revenue; in roofing, software is 0.22% of receipts.

  7. Not supported: the claim that Harvey is becoming a law firm. No source we opened shows it, and Harvey's own post says its commitment to law firms is not changing.

  8. Could not be sourced: three of the video's claims. Nvidia undercutting the AI labs, investors not knowing the ground, distribution not being a moat: we found no source for any of them.

The big claims about assistants and owning the service held. The 2% figure, the Harvey claim and two broad claims did not hold as stated.

What this shows

Labels on this scene

  • Picture Desk film: generated illustration · mood only.

Which lessons from the video survived the check?

The big claims about assistants and owning the service held. The 2% figure, the Harvey claim and two broad claims did not hold as stated.

You keep the lessons that held, and drop the ones that did not.

Grades from the paper's claim ledger: supported, partial or unsupported, checked against primary sources. The full ledger of 24 claims is downloadable from the paper.

  • Our reading Graded supported: general-purpose assistants already do much of what point solutions sold. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading Graded supported: becoming the service captures the budget that software never touches. General Catalyst states it as its thesis and is funding it. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading Graded partial: customers want one internal foundation, not a shelf of tools. Neither source measures what buyers want. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading Graded partial, too strong: each earlier era of AI tools is now obsolete. Plugins are a current ChatGPT feature again, and workflow automation is still sold. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading Graded partial: software is about 2% of a law firm's revenue. Clio measures 0.58% to 1.77% of expenses, and expenses are not revenue. The order of magnitude holds. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading Graded unsupported: Harvey is walking the path to becoming a law firm. No source shows Harvey operating as a law firm. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading Graded unsupported: Nvidia will undercut the frontier labs with in-house hardware and compute-only pricing; venture investors are still underwriting the old playbook and do not know the ground; distribution is not a moat. No source the paper opened establishes them. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading Foundation turned the video into 24 checkable claims and graded each against primary sources: 4 held as stated, 12 held in part, 8 could not be sourced and none was contradicted. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Official description November 2023: OpenAI's GPTs let anyone build custom assistants without code, and ChatGPT searches attached PDFs. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Official description November 2024: Anthropic launched the Model Context Protocol, one open protocol to connect assistants to business systems. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Official description October 2025: Anthropic shipped Agent Skills for Excel, PowerPoint, Word and fillable PDFs. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Official description By October 2026 OpenAI's plugins page invited business users to connect their apps "right into ChatGPT", from finance and sales to customer support and small business. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Attribution General Catalyst writes that technology builders capture more value by owning and operating the service businesses; it raised $1.5 billion in 2024 for that strategy, according to PitchBook. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading Follow one job from the first call to the last review. In the point-solution pitch each step gets its own tool, and each tool keeps its own copy of the homeowner. An owner who buys a tool per step becomes the integrator. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Official estimate In 2022 U.S. roofing contractors with employees reported $69.9 billion in receipts. They paid $13.5 billion in payroll and expensed $154 million of software: 0.22% of receipts, or 0.33% counting purchased computer services. Payroll was about 87 times the software line. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Attribution Harvey's chief executive told TechCrunch that about 4% of revenue came from corporates at the start of 2025 and 33% did by November. Its March 2026 post also says "Nothing about our commitment to law firms is changing." Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).

Download the claim ledger (CSV) →

Put a tool to the test.Follow the next part ↓

Put a tool to the test.

A tool survives if it has something an assistant cannot copy. Four things count.

Pick a tool. Does it have something an assistant cannot copy?

Put a tool to the test: what we found for each tool
ToolResultData the assistants cannot getA place in your workflow that is expensive to leaveA rule that requires itAn outcome it stands behind
AI phone receptionist0 of 4: fallsNot that we found. Phone calling is built into OpenAI's voice platform: calls since August 2025, $0.05 a minute for the voice session since September 2026, model and tool use extra.Ask the vendor. Only if it is part of the system where you run the business. Jobber already sells an AI receptionist inside its field-service suite.Not that we found. In our reading, the duty to tell callers they are talking to AI lands on whoever runs the line, whatever tool answers.Ask the vendor. Is it paid on your result, or a flat fee per user?
Insurance supplement writer0 of 4: fallsNot that we found. The price data is Verisk's: “We own all Price Data provided with the Services.” Verisk now brings its own tools into Claude.Ask the vendor. Only if it is part of the system where you run the business. Ask whether it keeps its own copy of your customers.Not that we found. Texas and Florida decide who may act for a policyholder on a claim. In our reading, those rules protect the licensed adjuster, not the software that drafts the letter.Ask the vendor. Is it paid on your result, or a flat fee per user?
Instant-quote tool0 of 4: fallsNot that we found. EagleView says it owns its aerial imagery and announced its own AI product in April 2026. Google's Solar API terms list energy uses, not roofing quotes.Ask the vendor. Only if it is part of the system where you run the business. Ask whether it keeps its own copy of your customers.Not that we found. In our reading, the Texas and Florida claim rules protect licensed people, not software.Ask the vendor. Is it paid on your result, or a flat fee per user?
The data owner's own AI (Verisk, EagleView)1 of 4: staysYes. In our view, a tool built on data its maker owns is not a wrapper. Verisk says it owns its price data; EagleView says it owns its aerial imagery. Both have announced their own AI.Ask the vendor. Only if it is part of the system where you run the business. Ask whether it keeps its own copy of your customers.Not that we found. In our reading, the Texas and Florida claim rules protect licensed people, not software.Ask the vendor. Is it paid on your result, or a flat fee per user?
Your job record (CRM and field system)1 of 4: staysAsk the vendor. It holds your own jobs and customers. Ask what happens to that data if you leave.Yes. In our reading, the system where you run the business is costly to leave. Sequoia wrote in 2023 that workflows hold up better than data.Not that we found. In our reading, the Texas and Florida claim rules protect licensed people, not software.Ask the vendor. Does the vendor stand behind a result? Contractor systems like Jobber and Contractors Cloud are already opening to the assistants.

The rule: a tool stays when it has at least 1 of the four. Otherwise it falls. “Ask the vendor” never counts as yes.

A tool you use, build or back Answer the four questions for it.

  1. Does it hold data the assistants cannot get?
  2. Is it where you run the business, so it would be costly to leave?
  3. Does a law or rule require this tool, not just a licensed person?
  4. Is the vendor paid only when you get the result?
What this shows

Your tool's result goes on your card.

Put a tool to the test: what we found for each tool
ToolResultData the assistants cannot getA place in your workflow that is expensive to leaveA rule that requires itAn outcome it stands behind
AI phone receptionist0 of 4: fallsNot that we found. Phone calling is built into OpenAI's voice platform: calls since August 2025, $0.05 a minute for the voice session since September 2026, model and tool use extra.Ask the vendor. Only if it is part of the system where you run the business. Jobber already sells an AI receptionist inside its field-service suite.Not that we found. In our reading, the duty to tell callers they are talking to AI lands on whoever runs the line, whatever tool answers.Ask the vendor. Is it paid on your result, or a flat fee per user?
Insurance supplement writer0 of 4: fallsNot that we found. The price data is Verisk's: “We own all Price Data provided with the Services.” Verisk now brings its own tools into Claude.Ask the vendor. Only if it is part of the system where you run the business. Ask whether it keeps its own copy of your customers.Not that we found. Texas and Florida decide who may act for a policyholder on a claim. In our reading, those rules protect the licensed adjuster, not the software that drafts the letter.Ask the vendor. Is it paid on your result, or a flat fee per user?
Instant-quote tool0 of 4: fallsNot that we found. EagleView says it owns its aerial imagery and announced its own AI product in April 2026. Google's Solar API terms list energy uses, not roofing quotes.Ask the vendor. Only if it is part of the system where you run the business. Ask whether it keeps its own copy of your customers.Not that we found. In our reading, the Texas and Florida claim rules protect licensed people, not software.Ask the vendor. Is it paid on your result, or a flat fee per user?
The data owner's own AI (Verisk, EagleView)1 of 4: staysYes. In our view, a tool built on data its maker owns is not a wrapper. Verisk says it owns its price data; EagleView says it owns its aerial imagery. Both have announced their own AI.Ask the vendor. Only if it is part of the system where you run the business. Ask whether it keeps its own copy of your customers.Not that we found. In our reading, the Texas and Florida claim rules protect licensed people, not software.Ask the vendor. Is it paid on your result, or a flat fee per user?
Your job record (CRM and field system)1 of 4: staysAsk the vendor. It holds your own jobs and customers. Ask what happens to that data if you leave.Yes. In our reading, the system where you run the business is costly to leave. Sequoia wrote in 2023 that workflows hold up better than data.Not that we found. In our reading, the Texas and Florida claim rules protect licensed people, not software.Ask the vendor. Does the vendor stand behind a result? Contractor systems like Jobber and Contractors Cloud are already opening to the assistants.
  • Our reading A point solution survives when it has something an assistant cannot copy, and four things count: data the assistants cannot get, a place in the workflow that is expensive to leave, a rule that requires it, and an outcome it stands behind. If a tool has none, the argument applies to it. If it has one or more, it does not. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading A tool built on a dataset it owns is not a wrapper. In the Verisk and EagleView cases, the tool is the data owner's own product. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading A system of record where the owner runs the business is a place in the workflow that is expensive to leave. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading Texas and Florida decide who may speak for a policyholder on a claim. Neither statute mentions software. Rules like these protect the licensed person, not the point solution that drafts the letter. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading A vendor paid on the result rather than the seat is aligned with its customer in a way a license cannot be. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading The roofing wedges pitched to Foundation most often are an AI voice receptionist, an insurance supplement writer and an instant-quote tool. They are real tasks with real money behind them, and some of them work. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Official description In August 2025 OpenAI's realtime voice model added phone calling over SIP. In September 2026 its newest voice model went generally available at $0.05 a minute for the voice session, with model and tool usage charged separately. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Official description Contractor systems of record are opening to the assistants: Contractors Cloud publishes a guide for connecting Claude or ChatGPT through MCP, and Jobber, which sells an AI receptionist inside its field-service suite, opened its data to ChatGPT and Claude in September 2026. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading In Foundation's reading, the duty to tell callers they are talking to AI lands on whoever runs the phone line, whatever tool answers it. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Official description Florida requires every party's consent to record a call; Texas requires one party's. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Official description Verisk's license says "We own all Price Data provided with the Services", and in May 2026 it announced connectors that bring its analytics into Claude. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading Four questions to put to any tool before you buy it: Does it write to your job record, or keep its own? What happens to your data and your workflows if you leave? Who answers when it is wrong, and what does that cost you? What does it do that the assistant you already pay for will not do next year? Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading Follow one job from the first call to the last review. In the point-solution pitch each step gets its own tool, and each tool keeps its own copy of the homeowner. An owner who buys a tool per step becomes the integrator. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Official description Texas bars a contractor from acting as a public adjuster on its own jobs, and Florida bars a licensed contractor from performing public adjuster services unless it is also licensed as one. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Official description EagleView states that "All aerial imagery is collected and owned by Eagleview", and in April 2026 it announced its own agentic AI product. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Official description Google's Solar API returns the dimensions of a building, but its terms list energy-system uses, not roofing quotes. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Our reading A roofing system where the owner actually runs the business, the CRM and field system of record, is not a point solution. It is a candidate foundation. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Attribution Sequoia warned in 2023 that data moats are "on shaky ground" and saw workflows and user networks building more durable advantage. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
One cause behind all three dead ends.Follow the next part ↓

Scene unavailable. The poster frame and the words in HTML

Row by row, paper workflow cards tip off the edge of a table and fall into a dark void; a gold-edged card falls last

A one-job tool becomes someone else's feature.

  1. A one-job tool becomes someone else's feature. The cause is the same for everyone: the budget is thin, and the big assistants take in the task.

  2. A roofing company spends about a third of 1% of its revenue on software and computer services. Split eight ways, one tool gets about $1,200 a year.

The budget is thin and the big assistants take in the task. What differs is who is left holding it.

What this shows

Labels on this scene

  • Picture Falling cards film: generated illustration · mood only.

What do the three dead ends have in common?

The budget is thin and the big assistants take in the task. What differs is who is left holding it.

You hold the one cause behind all three dead ends.

Eight tools is an assumption; the 0.33% is the U.S. Census figure for roofing companies with employees in 2022, counting software and computer services.

  • Our reading The cause is the same each time: the seat is thin and the wedge gets absorbed. What differs is who is left holding it. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Official estimate In 2022 U.S. roofing contractors with employees reported $69.9 billion in receipts. They paid $13.5 billion in payroll and expensed $154 million of software: 0.22% of receipts, or 0.33% counting purchased computer services. Payroll was about 87 times the software line. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Derived arithmetic At the measured share, spread across an assumed eight tools, the average roofing company spends about $1,200 a year on any one of them ($1,196). A vendor selling one tool at that price needs 8,360 customers to reach $10 million a year, 35% of all roofing firms with employees, and 83,595 to reach $100 million, 3.5 times the number of roofing firms with employees. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18); seat-arithmetic.
  • Our reading A wedge is a feature on someone else's roadmap. Tools fall. Outcomes stay. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
What lasts.Follow the next part ↓

Scene unavailable. The poster frame and the words in HTML

Wide shot at sunrise: a roofing crew carries a ladder and tool bags to the truck in front of a finished house

What lasts

What lasts: owning the outcome.

  1. What lasts is owning the outcome. In our view, three things gain value as AI improves, instead of being replaced: how a company runs the work, the records it leaves, and answering for the result.

  2. A foundation is all three, owned by someone who answers for the outcome.

  3. When the system cannot handle something, it goes to a person. Owners and crews are the escalation path.

  4. A tool earns a sliver of each job. Whoever owns the outcome keeps what builds up.

The workflows a company runs, the data they leave behind and the accountability for the result, owned by someone who answers for the outcome.

What this shows

Labels on this scene

  • Picture Sunrise film: generated illustration · mood only.

What is a foundation?

The workflows a company runs, the data they leave behind and the accountability for the result, owned by someone who answers for the outcome.

You know what compounds when the models improve.

  • Foundation position Three things compound when a model improves instead of replacing the product around it: the workflows a company runs, the data they leave behind and the accountability for the result. A foundation is the three together, owned by someone who answers for the outcome. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Foundation position Owning the outcome does not make the people doing the work a cost line. In a roofing company, owners and crews handle what the system escalates. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Foundation position A tool earns a sliver of each job. Whoever owns the outcome answers for it, and keeps what compounds. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Foundation position What lasts is owning the outcome: the workflows, the data they leave behind and the accountability for the result, with people handling what the system escalates. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
Your next step.Follow the next part ↓

Scene unavailable. The poster frame and the words in HTML

Golden hour: a roofer and a homeowner shake hands on the lawn in front of a house with a fresh roof

Your next step.

Every claim on this page comes from the research paper, which lists each source and its limits.

For Roofing company owners

Start a private conversationFor roofing company owners: talk through your tools and your system with us.Save my resultsYour tool test and the four questions

For Software builders

Start a private conversationBuilding for roofers? Compare notes with us.Save my resultsThe four tests and the sources

For Investors

Join the investor waitlistThe current round is closed. The waitlist sends one email if another round opens.Save my resultsYour question for the next pitch, with the sources

Read the research paper Watch the video we checked Download the claim ledger (CSV)

Films on this page are AI-generated illustrations, not footage of any company or product. Roofing figures are U.S. Census data, per-tool figures are arithmetic on an assumed eight tools (not a forecast), and Foundation, which builds a platform for roofing companies, is not neutral.

What this shows

Labels on this scene

  • Picture Closing film: generated illustration · mood only.
  • Disclosure Films on this page are AI-generated illustrations, not footage of any company or product. Roofing figures are U.S. Census data, per-tool figures are arithmetic on an assumed eight tools (not a forecast), and Foundation, which builds a platform for roofing companies, is not neutral.

What is the next step?

You leave with the four tests and a next step.

  • Foundation position This is research and opinion from an interested party. It is not investment, legal or insurance advice, and it is not an offer of securities. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).
  • Foundation position Foundation Projects is building a founder-aligned platform for roofing companies, so it is not neutral on this question. Source: A shelf of tools is not a foundation (paper, Markdown edition, read 2026-10-18).

Your card.

Your card · Roofing company owners

My tool check (owner)

The four tests and the four questions, in our view.

The four tests (our view)

  • Data the assistants cannot get
  • A place in your workflow that is expensive to leave
  • A rule that requires it
  • An outcome it stands behind

Ask before you buy

  • Does it write to my job record, or keep its own?
  • What happens to my data and workflows if I leave?
  • Who answers when it is wrong, and what does it cost me?
  • What does it do that my assistant will not do next year?

Sources

Every line comes from the research paper "A shelf of tools is not a foundation": foundationproject.net/research/ai-point-solutions-roofing. Its sources and the graded claim ledger are listed there.

Foundation's interest

Foundation Projects builds a platform for roofing companies, so we are not neutral. This is research and opinion, not investment, legal or insurance advice, and not an offer of securities.

Your answers stay in this browser. No form is submitted. They never go into a link.

Your card · Software builders

My moat check (builder)

The four tests, in our view.

The four tests (our view)

  • Data the assistants cannot get
  • A place in your workflow that is expensive to leave
  • A rule that requires it
  • An outcome it stands behind

After the wedge

A wedge buys an introduction. Then a vendor needs a place in the workflow that is expensive to leave, a network, or an outcome it stands behind.

Sources

Every line comes from the research paper "A shelf of tools is not a foundation": foundationproject.net/research/ai-point-solutions-roofing. Its sources and the graded claim ledger are listed there.

Foundation's interest

Foundation Projects builds a platform for roofing companies, so we are not neutral. This is research and opinion, not investment, legal or insurance advice, and not an offer of securities.

Your answers stay in this browser. No form is submitted. They never go into a link.

Your card · Investors

My pitch check (investor)

The four tests and the question to ask, in our view.

The four tests (our view)

  • Data the assistants cannot get
  • A place in your workflow that is expensive to leave
  • A rule that requires it
  • An outcome it stands behind

The ceiling

Software and computer services are 0.33% of roofing receipts (software alone 0.22%): about $9,570 a year for an average company. Split across eight tools (an assumption), about $1,200 each; $10 million a year then takes 35 of every 100 roofing companies as customers.

Sources

Every line comes from the research paper "A shelf of tools is not a foundation": foundationproject.net/research/ai-point-solutions-roofing. Its sources and the graded claim ledger are listed there.

Foundation's interest

Foundation Projects builds a platform for roofing companies, so we are not neutral. This is research and opinion, not investment, legal or insurance advice, and not an offer of securities.

Your answers stay in this browser. No form is submitted. They never go into a link.

Your choices and answers will appear here.

Kept in this browser. No central tracking or form submission.