Not subscribed? Sign up to get it in your inbox every week.

PRESENTED BY KOAH
Get paid for your AI app's free users

Most people using your AI app will never subscribe, and every message they send still lands on your inference bill.

Koah turns those conversations into revenue with ads built for AI apps, native to chat and matched in real time to what each user is asking.

AI apps like EcoGPT and Daimon already run it, and publishers earn 3x what legacy ad platforms pay.

Drop in the SDK and start earning revenue today.

Only using ChatGPT at work? You’re behind.

Most companies that "use AI" barely do. The gap between dabbling and doing decides who wins the next two years.

Every two weeks, the Census Bureau asks U.S. businesses one question: did you use AI in the last two weeks? The latest answer is about 24%.

Census Bureau BTOS: share of U.S. businesses using AI, current and expected Source: U.S. Census Bureau, Business Trends and Outlook Survey. Chart from census.gov/hfp/btos/data.

One in four sounds like progress until you look at what counts as a yes. A company where one person asks ChatGPT to draft an email checks the box. So does a company where agents enter orders, chase approvals, and close out invoices around the clock. They give the same answer on the survey but the scale of impact is completely different.

When I talked to the team at Hyperscale, whose AI specializes in back-office work, they put numbers on that gap. Teams that have fully leaned in see something like 20-to-1 returns. Teams that are dabbling break even at best. Adoption is a curve, and almost everyone is at the flat end.

Whether to adopt AI stopped being the question. This week is about getting from the flat end of that curve to the steep one.

Three things that changed this year

I've heard the same three points from operators and founders on stage all fall.

AI got 10x better this year. It will again next year. Things that broke a year ago work now. As Hyperscale cofounder Oliver Song put it, "Twelve months ago, this technology simply didn't exist at production quality." If you shelved an AI pilot in 2025, your conclusion has expired.

Connected systems are a must. AI can only work on what it can reach. If an agent can log into your CRM or ERP, inbox, accounting, and phones, it can take on the work that crosses those four systems. If it can only log into the inbox, it can answer email and nothing else, and your team is still copying numbers between the other three. (Side note: a system with no API or one that only exports to PDF is a reason to replace it. Continuing with an antiquated system is how you fall behind.)

"Our data is too messy" isn't a reason anymore. This used to be the honest blocker. Now an agent can pull from several sources, notice when they disagree, and decide whether to resolve it or pull a person in. Give it the ability to text, call, or email your team and it'll go get the missing piece the way a good coordinator would. It gets to the right answer the vast majority of the time, and when it isn't sure, it asks.

Where it started

Hyperscale's first deployment was deliberately simple. PS Logistics, who runs more than 5,000 trucks as one of the largest transportation companies in the US, had drivers sitting on hold for pickup numbers and repair updates. They put Vic, Hyperscale’s AI, on the phone line. PS's VP of IT, Mauricio Paredes, said it resolves roughly 70 to 75% of calls, and that it's "how we started our AI journey."

The work lives in the back office

Once an agent can read your systems, the questions change from "where's my load" to "why does this take three days and four people to complete." That's where the work actually is: in the seams between products, where someone copies a number from one screen into another.

The kind of thing teams now hand to an agent:

  • Short pays, traced to the cause. Every morning Vic compares what customers were billed to what they paid and sorts each gap by likely cause: an unposted credit memo, an unpaid surcharge, a charge that wasn't on the order. One customer's eight short pays traced to a single setup error: every invoice carried two fuel surcharges, and they paid one. Vic put the evidence in a table and drafted the fix for the rating team.

  • Warranty money nobody was looking for. Each morning Vic checks the previous day's repair work orders against the warranty and coverage plans on each unit and emails the warranty team the repairs that may be claimable. It also says plainly when there's nothing to claim, so nobody chases dead ends.

  • The recurring spreadsheet nobody wants to rebuild. An operations analyst needed a workbook of every repair order and purchase order matched against its vendor invoices, with remit-to details added and mismatches flagged. Vic reads the invoices, rebuilds the whole workbook on request, and flags the ones it couldn't read cleanly for a person to check.

  • A PDF estimate that becomes a finished work order. A repair estimate that arrives as a PDF becomes a finished work order, with line items entered, vehicle and driver filled in, and the total checked against the estimate.

Top-down vs. groundswell

Companies usually roll out AI in a few ways:

Top-down. Leadership picks the use cases, buys the tools, and sets the targets. You get focus, budget, and real connections to your systems of record. You also get a list written by the people furthest from the work. The use cases are reasonable but overly focused on a few high volume use cases, which in reality only account for a portion of total work.

Groundswell. You give everyone access and let them figure it out. You get creativity: the scheduler who automates her own Monday report, the AP clerk who builds a vendor follow-up nobody asked for. You also get a hundred disconnected experiments, no access to shared systems, and nothing that compounds.

The companies at the 20-to-1 end do both at once. Leadership connects the systems and sets the guardrails. The whole team decides what to build on top and iteratively bubbles up high value use cases.

When it works, AI stops being a project and becomes integral to how your team solves problems while eliminating manual, repetitive work. The people closest to the work end up designing it.

How Hyperscale runs both at once

Most of the stalled AI rollouts I hear of have the same problem: people don’t know what they’re missing. You’d have to break your own job down into tasks first, and most people have never done that.

What struck me about Hyperscale is that they treat this as the product’s job, not the customer’s. You describe how the work gets done, and it finds the repetition on its own. They do it from two directions:

Top-down. Vic connects to the systems a business already runs (e.g., CRM or ERP, inbox, accounting, phones, etc.) and maps where work actually moves: which tasks get touched over and over, where information gets copied between systems, and where things sit waiting on a person. Leadership gets a map of where AI will pay off, built from how the business really runs rather than how the org chart says it does.

Groundswell. Everyone on the team gets Vic too. It learns how each person works and, instead of waiting to be asked, suggests things it could take off their plate, then builds them with that person. Nobody sits through AI training. People learn it on the job, one suggestion at a time.

If you want to see what I’m talking about for yourself, Hyperscale is running an early access program until January 1, 2027. They're building the clearest picture I've seen of how agents get into real-world businesses.

Till next time,
Chris

Would you share with a friend?

Login or Subscribe to participate

Reply

Avatar

or to participate