The scam is that it works

The diagnosis

Why the agency that delivered still left you where you were

The obvious scam is real and boring. In June 2025 Gartner estimated that of the thousands of companies selling agentic AI, only about 130 were real, and coined agent washing for the rest. You do not need an essay to spot that one.

The one worth writing about is the agency that delivers. The workflow has fired at 9am every morning for four months. There is a walkthrough video and an invoice that got paid. Nobody lied to you, and the business is exactly where it was, because nobody checked whether the thing being automated was the constraint.

Here is what matters:

• •Gartner surveyed 350 executives at billion-dollar companies already running AI agents. Around 80% had cut headcount, and the rate of cutting was nearly equal between the firms reporting strong ROI and the ones reporting modest or negative outcomes. The automation landed. The outcome did not follow.
• •A randomized trial in Indian textile plants lifted productivity 17% in the first year using free consulting on standard practice. Just over 45% of the initial non-adoption came from owners who had heard of the practice and judged it would not pay in their business. That is not ignorance. It is a wrong theory about your own company, held confidently, by the person best placed to know better.
• •An audit attempted a web form submission at 14,061 companies. Of the 9,538 that had a working form and received one, 4,472 never responded at all. Ask those businesses their bottleneck and most will say lead generation.
• •The two statistics you will be sold from slide four do not survive being opened. RAND's 80% of AI projects fail is hedged in the report itself as by some estimates and endnoted to a 2022 magazine article. MIT's 95% of pilots return nothing is a non-peer-reviewed working paper that uses three different denominators for the same number.
• •Three questions before you sign anything. If this worked perfectly tomorrow, what breaks next. Where does the money actually leak, by numbers rather than instinct. And is this a technology problem, or a decision you have been avoiding.

The principle

What was the affordability check actually for?

Here is a question that sounds simple and quietly dismantles modern lending. If a bank verifies your income, your employment and your history, concludes you can repay, and then writes a contract where you carry 100% of the downside, what exactly was the affordability check for?

The sharpest objection I get is that risk sharing works for startups because a startup generates profit to share, while a house does not. It is reasonable-sounding and exactly backwards. Risk sharing is not about the asset earning money. It is about the financier sharing the asset's downside so its incentives point the same way as yours.

Here is what matters:

• •Mian and Sufi's own example: buy a $100,000 home with an $80,000 mortgage and your equity is $20,000. Prices fall 20% and you lose all $20,000 while the lender is untouched. Debt concentrates the first loss on the least cushioned party by design, not by accident.
• •The same economists proposed the shared-responsibility mortgage: if local house prices fall 20%, your principal falls 20%, and in exchange the lender takes 5% of any capital gain. Downside shared, upside shared.
• •Nobel laureate Robert Shiller proposed the continuous workout mortgage, which indexes the balance to local prices so nobody has to foreclose to trigger a workout. He has spent years asking why housing finance is still stuck at such a primitive stage.
• •This is not a religious argument. Islamic finance builds on it through diminishing musharaka, where loss follows ownership, but so does venture capital, so does leasing, and so do income share agreements. Different traditions arrived at the same structure independently because it works.
• •The incentive is the whole point. A lender fully protected by collateral finds foreclosure rational even when a workout would serve everyone better. A financier holding part of your downside would rather keep you solvent.

New in the directory

458 organizations, and a third of them do not want your equity

Where it stands today: 458 organizations across 59 countries and 10 macro-regions, of which 169 are non-dilutive or program-based, and 22 are explicitly halal or Sharia-aligned across 7 countries. Recounted from the data file on the day this went out, not copied from an old snapshot.

Honest note on additions: this was a thin month. One organization was added on 28 July, Silkroad Innovation Hub. The change that mattered was structural: 39 organizations now surface in the regions they genuinely serve rather than only where they are headquartered, so a founder in Central Asia searching their own region finds the Palo Alto fund that exists to back them. From the next issue I will name every organization added, with the date.

So instead of padding the list, here are three programs that take no equity and were confirmed open when this was written:

• SBIR/STTR, United States →

  Phase I up to roughly $323K, Phase II up to roughly $2.15M, across 11 federal agencies. Their own site is blunt about the terms: 0% equity or IP ownership taken by the government, around $4 billion invested a year across roughly 4,000 companies.

• Innovate UK →

  Grant competitions on fixed windows, not rolling. Open as of writing: Farming Futures automation and robotics, a share of up to £20 million, closing 30 September 2026. The UK to Singapore collaborative R&D round, up to £3 million, opens 24 August 2026.

• Astana Hub, Kazakhstan →

  Three programs listed as accepting applications, including a Central Eurasia market entry track for international B2B startups. Participants get 0% corporate income tax status. Useful if you are looking at a region most directories ignore.

Receipts

What actually happened this month

• Selected

  Taxformify is one of Edmonton Unlimited's 10 Companies to Watch for 2026, chosen from 59 applicants and judged by Ashif Mawji, Yasmine Al-Hussein and Aroon Sequeira. To be exact about what that is: a shortlist and a pitch on 5 October, not a prize already won.

• Covered

  BetaKit reported the selection and Taproot Edmonton relayed our claim that the software saves 30 hours a month per 100 clients. That is our number, not their finding, and I would rather you hear that from me than assume a journalist verified it.

• What is not working

  Search. In the three months to 30 July this site had 2,321 impressions and 9 clicks in Google Search Console, at an average position around 23, and zero searches for my own name. I write about earned rather than borrowed attention, so you should know that mine is not earned yet. The essays rank on page two and three, where nobody clicks. That is the problem I am working on, and you will see the number again next month whether it moved or not.

One question before you go

What is the thing you were about to buy to fix a problem you have not measured?

Hit reply and tell me. I read every one, and I answer.