What I’m thinking about: Why trust in AI keeps sliding even as the models get scary good, why ~95% of companies still see no return on it… and the one habit separating the operators pulling away from the ones drowning in “work slop.”
Last week I broke down where AI in business is heading: Marc Benioff, Salesforce’s ~$300M Anthropic token bill, and the durable moat for value-focused operators.
This week is the uncomfortable other half. Less about where AI is going… more about why most people (and companies) are falling further behind.
=====
This recent Gallup poll in regard to Gen Z and AI fascinated me.
Bottom line, AI usage is essentially flat in that cohort, though positive sentiment is sharply down.
This specific poll result was most encouraging to me though, regardless of the increase in negativity:
Get folks using AI, and they will feel more empowered. (Is this not a version of Dan Sullivan’s and Ben Hardy’s The Gap and The Gain?)
Nevertheless, the frontier labs haven’t helped their case. When your headline message is “a lot of white-collar jobs (maybe all of them, but especially the entry-level ones) are going away,” you shouldn’t be shocked that few people line up to cheer you on.
Even so, it’s hard for me to contemplate the decreased belief folks have in the practical capabilities of AI, even as the 2026 models blow past anything we had twelve months ago in terms of horsepower and reliable (and diverse) output (keep in mind this poll took place after the release of agentic tools like Claude Cowork):
Most folks (and companies) are using AI poorly, so of course it isn’t working for them
The above poll certainly isn’t the only one demonstrating negative sentiment in regard to AI. Search around and you’ll find plenty, like the widely-cited MIT study from last summer finding ~95% of enterprise AI pilots showed no measurable P&L return.
My read: these skeptical people aren’t wrong about their experience…they’re just blaming the wrong thing.
Many companies and individuals use AI badly: deprecated or free models months behind the frontier (any study conducted prior to the start of 2026 has to be heavily discounted, given the step-change increase in AI model capabilities since then), constrained or counterproductive techniques, heavy firewalls or restrictions from the IT department, or “work slop” being traded back and forth (more on that below). The productivity bump never shows up because the setup was broken from the start.
While we are by no means perfect, at SLC, we run the opposite way. Small shop, no internal red tape, AI-first 100% of the time, on the best models money can access, learning from the sharpest AI architects in the world.
And I’ve yet to find an industry AI can’t dramatically boost. I have an AI strategy call later this week with a groundskeeping business, about as physical as it gets, and there are STILL 10+ practical, revenue-driving AI implementations that immediately come to mind.
AI optimist with a clear-eyed ledger
I’m about as bullish on AI as anyone you’ll meet, though I won’t pretend the downsides aren’t real: AI-avatars stunting social-skill development, people of all ages spending less time together in person, wealth bending toward a handful of companies, and AI slop.
Not hand-waving any of those (major societal) concerns away, though the pros still outweigh the cons IMO.
At SLC, we are undeniably making progress from a hard ROI perspective (e.g. the hours we can count cutting out of a task, or decreased overhead from efficient AI workflows), in addition to soft ROI (the harder-to-measure capacity we’re building for future growth).
The soft kind matters more over a long enough runway (years to decades, though any prediction beyond ~2 years in this environment has to be made with extreme humility)…which is exactly why bean counters chasing this quarter’s number miss it.
=====
An additional point potentially contributing to AI skepticism is that, counterintuitively, the learning curve for AI got steeper, not easier…even as the models became leaps and bounds more robust and reliable.
In 2022, ChatGPT was a chat box. One door.
Now “AI” encompasses competing models with varying capabilities (e.g. ChatGPT, Gemini, Claude), Cowork versus Claude Code, projects, connectors, plugins, skills, and MCP hooks (the plumbing that lets AI plug into your Notion, Gmail, or Drive). The ceiling for output is FAR higher…and so is the on-ramp.
We’ve seen this movie. The internet went from dial-up and static pages to social platforms, SEO, complex ad-based marketing, varied e-commerce platforms (Amazon-only sellers, drop-shippers wired into global supply chains), and eventually crypto rails.
Every layer multiplied what you could build… and made the curve harder to climb, not softer. I see this serving as an AI Advisor inside Callan Faulkner’s Accelerate to Automate program. Substantial financial investment to get in, motivated people, and the client questions like, “Should I do THIS…or perhaps THIS…or maybe THAT?” are endless.
That being said, there’s never been a more forgiving learning curve, even if it is steep. AI can infinitely abstract and ‘dumb-down’ explanations for you, and it never loses patience. If you ask the right questions…you WILL get the right answer, though many times you will have to sift out the golden nuggets, and be willing to push back on AI responses (unfortunately, most folks just don’t trust themselves, on either side of the equation).
“Work slop”: the quiet reason the gains disappear
One of my mentors, Ryan Levesque of Ask Method, flagged a concept from a Harvard Business Review study: work slop.
You already know AI slop, the throwaway images and videos (…now business organic content and ads?) clogging every feed. Work slop is the professional cousin.
Someone means well, fires off a prompt, and ships whatever the model produces without ever checking it (encapsulated by my favorite analogy in this space: AI passenger vs the AI driver.)
Critically, work slop pushes the work downstream, so whoever receives it (usually the capable person whose name is on the outcome, or who is responsible for money changing hands) has to decode it, correct it, or rebuild it (I’d bet everyone reading this has dealt with this scenario by now).
AI can (and SHOULD) enhance productivity just fine, but work slop quietly eats the gains before they reach the bottom line.
We’ve built this ‘anti-work slop’ mentality into our core values at SLC: tech amplifies expertise, it never replaces judgment.
To be clear, I’m not preaching ‘read every line of output every time’. Once you’ve QA’d a process to ~95%+ confidence (e.g. producing our SEO and GEO content), you trust it without re-reading (though a quick set of eyes should still be in the routine).
The $10M parcel that will never get a reply
The second I read about ‘work slop’, it clicked, because I watch it land in our own inbox every week.
For example, someone recently sent us a parcel that “ChatGPT says is worth ~$10M because it sits in an industrial development pathway,” with a few bullet points of overly optimistic and generalized rationale, and zero grasp of the local market mechanics or how you’d dispo it in this market.
That doesn’t get a response from us. Cleaning up someone’s work slop isn’t our job, and nobody is pulling a ~$5M check out of us with a ChatGPT printout.
Show me the comps (and ideally a highly specified dispo pathway leveraging established local connections), or don’t send it.
There’s a quiet upside here, and it’s a big one. While most companies drown in their own self-induced slop, the operators who refuse to add to the pile stand out by default.
To reiterate, being an AI architect comes down to two things: asking the right questions to the model, and knowing what to keep.
Do that, and it carries your business further than you thought, in less time than you’d expect. That’s why we keep honing that skillset EVERY day.
=====
On the subject of what TO send us: our sharpest edge is still underwriting and capital. AI is just the leverage we use to deliver it faster.
If you’re an experienced operator with routine deal flow (the underwritten kind, not a ChatGPT printout), let’s build something that repeats.
We write checks from $50K+, we close 100% of the deals we commit to, and we bring national underwriting built across thousands of reviewed transactions.
When we commit to funding, we close.
