The Fax Machine Test: How to Tell If Your AI Vendor Is Real

Your inbox has three cold emails this week from "AI Strategy" firms you've never heard of. Your LinkedIn feed is wall-to-wall thought leadership. The big consultancies you already pay are talking about agentic transformation. Your local Facebook group has a guy offering AI workshops out of a co-working space.

A lot of it doesn't hold up.

Same pattern as 1995, when half the country was selling "Internet Strategy" to businesses and most of them couldn't tell you what an HTML tag was. Same movie, rebooted. The technology is real. The hype on top of it mostly isn't. The gap is where you get separated from real budgets for things that don't offer tangible value.

So here's a small checklist. Six tests. None of them require you to be technical. All of them work on websites, decks, cold emails, conference talks, your own internal pitches. Run them on anyone selling you something with "AI" in the name. Run them on me, too.

1. The Fax Machine Test

Read an AI offering. Anything. From a website, a deck, a cold email, a vendor pitch:

"We help mid-market businesses unlock the transformative power of artificial intelligence. Our proprietary framework delivers end-to-end AI transformation, accelerating value creation through agentic capabilities tailored to your unique business needs. Schedule your free discovery call today and start seeing measurable results within 60-90 days."

Now strip "AI" out and put a different boring tool in its place. Fax machine. Email. Web browser. Anything that's just a thing people use:

"We help mid-market businesses unlock the transformative power of fax machines. Our proprietary framework delivers end-to-end fax machine transformation, accelerating value creation through advanced fax capabilities tailored to your unique business needs. Schedule your free discovery call today and start seeing measurable results within 60-90 days."

If the swap reads as absurd, the original was too. You just didn't notice because "AI" gave it cover.

Nobody has ever paid for Fax Machine Strategy. Fax machines are tools. You buy one, plug it in, and use it. Anyone selling Fax Machine Transformation is either selling you nothing or doesn't know what a fax machine is. Both should disqualify them.

Real AI offerings tell you exactly what work they do - the actual workflow they touch, in the actual job a human used to spend Tuesday afternoon on. "We rewrite your contract review process so the first pass is automated." "We build an internal tool that reads your support tickets and routes them to the right team." That's a thing. The fax-machine version still makes sense: "We rewrite your fax intake process so the first pass is automated." Boring. Specific. Real.

2. Look for the noun

The Fax Machine Test catches the worst of the boilerplate. The next test catches what's left.

A real offering tells you the noun it acts on. A workflow. A document. A decision. A process step. A team's daily grind.

A hype offering points at something abstract. "Value." "Outcomes." "Transformation." "Capabilities." Words that could attach to anything because they don't actually attach to anything.

When you read a pitch, hunt for the noun. If it's a real noun - contract review, support ticket triage, invoice reconciliation, first-pass legal redlines - there's a decent chance someone has done some real work. If the noun is "value" or "growth" or "your business," there's nothing under it yet.

This isn't a 100% filter. Real offerings sometimes hide their nouns under marketing copy. But the inverse holds: an offering with no real noun anywhere in it almost never has real work behind it. If a marketing team writes vague copy, it's usually because vague is all they were given.

3. Beware the quantity claim

We're early. AI tradecraft is months old, not years. The patterns are forming in real time. Anyone who has built something real in this period knows how much one good thing took to build, and how much time the post-launch tuning takes.

So when you see a vendor leading with big numbers - hundreds of pre-built agents, dozens of ready-to-deploy use cases, a sweeping library of enterprise AI patterns - be skeptical. Not because the numbers are impossible, but because at this stage of the technology, quantity is usually a sign of un-quality. Either those things are mostly thin wrappers around the same prompt, or someone counted very generously, or the bar for what counts as an "agent" was set low enough that most of them wouldn't survive contact with a real workflow.

The real practitioners I've seen - the ones whose work makes me want to call them and ask questions - are usually working on a small number of things very deeply. A handful of workflows. A few agents. One internal system that's been refined across hundreds of sessions. The depth-to-breadth ratio is the tell.

A vendor with a thousand agents has a thousand demos. A real practitioner has three things that work.

4. Treat "autonomous" as a yellow flag

If you hear that someone's AI is making decisions on its own, running workflows on its own, operating without supervision - pause.

Autonomous AI is real, in narrow forms. But running an LLM-based agent on consequential decisions, without tight constraints around what it can and can't do, is either a research bet or a way to lose money fast. The tradecraft for keeping these things on rails - the monitoring, the feedback loops, the context management, the guardrails, the eval harnesses - is being figured out right now. Most teams claiming autonomy don't have most of it yet.

So when someone tells you they have an autonomous agent doing something serious, ask them what the rails look like. How does it know when it's wrong? Who reviews its decisions? What happens when it makes a bad call? What's the fallback? If those questions get vague answers, the "autonomy" is either marketing copy on top of something much simpler underneath, or a real risk to whatever business it's running in.

The best AI work right now is constrained AI work. Tight context. Clear scope. Human in the loop for anything that matters. The people doing it well over-describe their guardrails because the guardrails are most of the engineering.

5. The expertise should pre-date the AI

This one is the most useful and the easiest to verify. Look at the LinkedIn profile, the company history, the case studies. The expertise should pre-date the AI marketing.

If they were helping people with sales strategy in 2022, and now they're helping people use AI to boost sales strategy - that's real. The domain knowledge is the foundation; AI is the accelerant. If they were doing operational transformation work for a decade, and now they're doing AI-enhanced operational transformation, you're getting a decade of operational expertise plus a new tool. That compounds.

But if they appeared in early 2023 with a new title and have been an "AI Transformation Strategist" ever since - what was the strategy work before? Almost nothing? That's the tell.

AI doesn't substitute for expertise. It doesn't create expertise. It approximates expertise, the way a brilliant intern with a search engine can approximate it for an afternoon. And approximation is useful - but it's not a thing you can build a consulting practice on, because when the work gets hard the approximation breaks down and there's no underlying expertise to fall back on.

A consultant whose only credential is "I use AI" is selling you the AI's outputs. You can buy those outputs from the AI directly for $20 a month.

6. The timing test

Real AI work, applied to a specific operational workflow, shows results in days. Not 60-90 days. Days.

Think about it. If a vendor was deploying "Email Adoption Frameworks" in 1996, you wouldn't accept "you'll see measurable results in 60-90 days." Of course you wouldn't. Email is a tool. Either people start using it, or they don't, and you find that out by Friday.

AI applied to a real workflow is the same. You install a contract review assistant on Monday, the legal team uses it on a real contract Tuesday, and by Wednesday you know whether it saved them three hours or wasted them two. Days. Not quarters.

When you see "60-90 days to measurable results," what's actually being sold is a transformation engagement. A roadmap. A discovery phase. Multiple workshops. Stakeholder alignment. By day 90 you may have a slide deck and an internal champion and a pilot scoped - but you haven't done the work yet. The work hasn't even started.

The shortest path to knowing whether AI helps your business is to put it in front of one workflow and one team for a week. Not to commission a strategy.

What to ask for

A few things that should be in any real AI offering, that fakers can't easily fake:

• A specific workflow they've improved. With before-and-after numbers that aren't "revenue." Time saved per task. Errors caught. Cycle time. Things you can verify.
• A live demonstration with real things. Real practitioners can show you the work happening, in front of you, on real data - or, in AI's case, on test data they can simulate in minutes. Fakers want a 12-week onboarding before anything ships.
• Their own use of AI on themselves. If they're selling you a service their own company could benefit from, and they can't give you firsthand accounts, named use cases, and full transparency into how they've applied it internally - how good could the thing they're selling actually be?

A few things they can reasonably keep vague:

• The names of their clients. NDAs are real. "We work with three Fortune 500 healthcare companies" is fine if the rest of the pitch is concrete.
• The exact methodology. Real shops have proprietary process. They don't have to publish it. But they should be able to describe what the engagement looks like in plain English.
• The model or vendor stack. Plenty of good shops are model-agnostic. The model isn't the product; the work is.

The 1995 thing one more time

I built websites in 1995. There were three kinds of people in that market.

There were the people who'd been writing software for years and saw the web as a new place to ship it. Their work was good then and good now.

There were the people learning the new tool because they were curious and built things alongside their day jobs and got useful. Some of them turned into the next generation of real builders. Most stayed amateur and that was fine.

And there were the people who learned the buzzwords and sold "Internet Strategy" to companies that didn't know any better. They sold consulting engagements that delivered nothing. They built Flash sites for businesses that needed a contact form. They billed for six months of "discovery" and disappeared. By 2001 they were gone.

We're in the same window. The first kind are out there building real things. The second kind are learning in public, often noisily, sometimes well. And the third kind are setting up shop right now and flooding the market, social media, and our inboxes with empty calories.

Run the tests. You'll find the first kind doing the work.

Run them on me, too. I'm the second kind, with experience across a few of the disciplines that matter here, and I've done this before.