The Referee Never Wins
Only true executors will read (8 mins)
You don’t have an AI problem. You have an aiming problem.
I watched a plumber last week spend three hours making an AI-generated cinematic video of his van pulling up to a job site.
Dramatic lighting. Slow-motion wrench grab. The whole thing looked like a Marvel trailer for copper pipes.
It got 23 likes. I wonder how many calls he got.
Did you expect ‘Better Call Saul’ kind of situation to happen..? And look, I get the excitement, Something that used to require a full production crew and a five-figure budget can now be done on your laptop in an afternoon.
That feels powerful. It feels like progress.
But here’s what I keep seeing, over and over, from business owners, creators, and solopreneurs who are genuinely trying to use AI:
They’re not failing to use it. They’re using it on the wrong things.
Where AI Should Actually Be Pointed
The plumber doesn’t need a cinematic reel. He needs AI analyzing which neighborhoods have the oldest pipes and targeting those homeowners with a direct mail campaign.
The lawyer doesn’t need AI-generated headshots. She needs AI reviewing her last 50 client intakes to find the pattern in which leads actually convert and which ones waste her time.
The electrician doesn’t need a 4K image mashup for Instagram. He needs AI drafting follow-up sequences for every quote he sends out that never gets a response.
People know how to use AI.
That’s not the gap anymore. The gap is knowing where to point it.
The ROI Test Nobody’s Running
Here’s the question I think we should all be asking before we touch any AI tool:
Does this generate income or bring more value to someone?
Not “is this cool?” Not “can AI do this now?” Not “will this get engagement?”
Does it move the needle on revenue, efficiency, or the quality of what you deliver to the people who pay you?
When you run that filter, most of what people use AI for falls apart.
The excitement that what once required a specialist can now be done with a prompt is real. But that doesn’t mean those things nourish your business.
The capability is not the same as the use case. They’re disconnected.
Hoffman Sees It Too (From the Other Side)
Reid Hoffman made a similar point in a recent conversation, though from the opposite direction.
He said that if you’re not finding AI useful for substantive work, research, decision support, information analysis, you’re not trying hard enough. He uses AI to do deep research when writing books. He sets queries running for 10 to 15 minutes of compute and comes back to synthesized answers that would have taken an analyst a full day.
He’s right. But I’d push it further.
Most people are trying. They’re just excited about the wrong things.
They see AI can generate a video, so they generate a video. They see AI can make an image, so they make an image.
The novelty is real. But novelty doesn’t pay rent.
The highest-value skill in 2026 isn’t using AI. It’s asking AI the right questions.
Thinking about one thing. Thinking about another. And having AI connect the two into something bigger.
That’s a different sport entirely. And almost nobody is playing it.
Why Everyone’s Aiming at the Wrong Target
There’s a reason for the misalignment. Hoffman identified it clearly, even though he was talking about big companies.
Most organizations, and most individuals, start from a risk-first mindset. Avoid the downside. Gain the upside second.
In practice, that means: don’t try anything real until you’ve eliminated every possible risk.
What if the AI makes a mistake in a client proposal? What if it writes something inaccurate? What if the output sounds robotic?
So instead of pointing AI at the work that actually matters (lead generation, research, decision-making, follow-ups), people default to the safe stuff.
The stuff where a mistake doesn’t matter:
Make me a pretty picture.
Write me a social media caption.
Generate a fun video.
That’s the compliance officer mindset at the individual level.
You’re not avoiding AI. You’re avoiding using it where it counts because the stakes feel higher there.
The Driveway Problem
Hoffman compared it to refusing to drive until you’ve eliminated every risk on the road.
You’ll never leave the driveway.
And what happens when you never leave the driveway is you spend all your fuel idling. You’re using AI, burning time and energy on it, but you’re going nowhere.
The ROI isn’t just low. It’s negative.
The fix isn’t “use AI more.”
The fix is to aim it at the things that scare you a little. The client-facing work. The revenue-generating processes. The decisions you’re currently making on gut instinct that could be informed by 15 minutes of AI research instead.
That’s where the value is. Everything else is decoration.
Get on the Pitch (Or Stay a Referee Forever)
Hoffman tells a story about Europe that I think applies to every business owner sitting on the sidelines.
He says Europe’s default posture in the AI race is to be the referee. Writing regulations, setting standards, defining what’s allowed.
His response is blunt: the referee never wins, and nobody likes the referee.
You have to get on the pitch. You have to play.
I think this is exactly why even well-funded European AI companies struggle to keep up with US competitors. Mistral AI, one of France’s best AI startups, is genuinely impressive. But they’re operating inside a regulatory environment so tight that even the best players are running with weights on.
Meanwhile, US companies are iterating at full speed, breaking things, fixing things, and shipping.
It’s the same dynamic at the small business level. The owner who says “I need to fully understand AI before I use it on anything important” is the referee. The owner down the street who just started using it on follow-ups last Tuesday and already closed two extra jobs? That’s the player.
This Scales Down to You
If you’re reading articles about AI but not using it on real work, you’re being the referee.
If you’re forming opinions about what AI can and can’t do based on what you’ve heard, not what you’ve tested, you’re being the referee.
If you’re waiting until AI is “good enough” or “safe enough” to deploy in your business, you’re being the referee.
Getting on the pitch means picking one real, revenue-adjacent process in your business and running AI on it this week.
Not as an experiment. As a commitment.
Record your next client call and run the transcript through AI for follow-up actions.
Use AI to research your top 10 competitors and identify what they’re offering that you’re not.
Let it analyze your pricing against your local market and tell you where you’re leaving money on the table.
The people doing this right now, while it still feels early and a little awkward, are building muscle memory that won’t be optional in two years.
The Contrarian Bet That Has Physics on Its Side
Hoffman identified a pattern across every great investment he’s ever made: LinkedIn, Airbnb, Facebook.
In each case, he could see why smart people thought it was a bad idea. And he could articulate why he thought they were wrong.
Being contrarian isn’t enough, though. You also have to be right.
And the only way to find out is by doing the work, not theorizing.
I have my own contrarian bet, and it’s this:
Local open-source AI is the future.
Here’s the Logic
As frontier models keep improving, open-source models improve right behind them.
The gap isn’t closing because open-source got lucky. It’s closing because the underlying research, the architectures, the training techniques, they all flow downstream eventually.
The frontier models that feel untouchable today will be matched by open-source alternatives in the near future.
And when that happens, even if frontier models remain technically superior, the math shifts:
Cost: If something can bring similar value at a fraction of the cost, or zero cost, people will take advantage of it. That’s just economics.
Privacy: From a security perspective, local models win by default. Your data never leaves your machine. No API calls. No terms of service. No third party holding your client information.
That’s why I think hardware costs are going to rise in the long run.
If you own solid hardware with high RAM and good processors, you’re set. The subscription model for AI won’t disappear. But the people who invest in running capable models locally will have an edge in cost, privacy, and independence.
Put it this way: the $1,500 you spend on a solid desktop today could replace $200/month in AI subscriptions within a couple of years. And your client data stays on your machine, not on someone else’s server.
Smart People Will Say I’m Wrong
They’ll say frontier models will always be too far ahead. They’ll say open-source can’t compete on quality. They’ll say local compute can’t match cloud scale.
Maybe.
But if you’re a business owner, here’s why this matters to you right now: the tools you’re paying monthly for today are going to have free, local equivalents soon. The question is whether you’ll be ready to use them, or whether you’ll be starting from scratch while your competitors already have the workflows built.
The pattern Hoffman identified says the contrarian bet is exactly where the outsized returns live.
And this one has physics on its side, not just market optimism. Models get smaller and more efficient over time. Hardware gets cheaper. The trend line is clear.
I’d rather own the hardware than rent the intelligence.
Edge Cases (Where This Might Not Land)
Heavily regulated industries (healthcare, finance, legal): The compliance concerns around AI in client-facing work are more legitimate. You can’t “move fast” with patient data. But even here, the answer isn’t “don’t use AI.” It’s “find the use cases where regulatory risk is low and start there.” Internal research. Meeting summaries. Competitive analysis. There’s always a starting point.
Hyper-local, relationship-driven businesses (small-town accountant, family lawyer): Using AI behind the scenes to be more prepared, more responsive, and more thorough doesn’t threaten the human relationship. It strengthens it. Your clients don’t need to know you used AI to research their situation before the meeting. They just notice you’re better prepared.
“I tried it and it didn’t work”: Check the prompt before you check the tool. A vague ask gets a vague answer. A specific, detailed prompt with context about your business, your client, and your goal gets something you can actually use. The gap between those two experiences is the whole game.
Action Checklist (This Week, Not Next Month)
Run the ROI test on your current AI usage. List everything you’ve used AI for in the last 30 days. Next to each item, write whether it generated revenue, saved meaningful time, or improved something a client would notice. If the answer is “no” on most of them, you’ve found the problem.
Pick one revenue-adjacent process and aim AI at it. Proposals. Follow-ups. Pricing research. Lead qualification. Client research before a meeting. Pick the one that makes you slightly uncomfortable and do it.
Record your next call or meeting and run the transcript through AI. Ask for action items, missed opportunities, and follow-up suggestions. See if it catches something you didn’t.
Ask AI a connecting question. Not “write me an email.” Instead: “I run a [type of business] in [location]. My biggest challenge is [X] and my goal is [Y]. What are three approaches I haven’t considered that connect these two things?” That’s where the real value lives.
Research one open-source AI model you could run locally. You don’t have to switch today. But understanding what’s possible without a subscription, without sending your data to a third party, is worth an hour of your time. The landscape is moving faster than most people realize.
Reid Hoffman has been wrong before. He says so himself. Some of his bets failed for the exact same reasons his wins succeeded.
But here’s what I know from my own work: the business owners who are winning with AI right now aren’t the ones who use it the most. They’re the ones who aim it at the right things.
They skip the spectacle and go straight to the work that pays.
The referee never wins. Get on the pitch. But when you get there, make sure you’re kicking toward the right goal.
Cheers,
your Jinni