Every week, someone asks me which AI tool to buy or use. Almost nobody asks me about the harness. I’m not saying they should today but this is about to become one of the most important questions when it comes to AI and AI adoption in enterprises.
It’s backwards that no one asks, and it’s the reason I wrote this. The tool question is comfortable. It comes with a shortlist, a demo, a vendor who walks you through the features with a slide ready for every objection you might raise. The harness question is uncomfortable, because it was never about the tool. It’s about what happens to the tool the moment it lands on someone’s desk.
We keep shopping for the engine and skipping the ride.
Here’s what that actually looks and feels like in your day to day.
You open Gmail because that’s where the note landed. You copy three numbers into Excel because that’s still where the spreadsheet lives. You drop a screenshot into a chat so someone else can see it, dig up last year’s PowerPoint because someone is going to ask for the slide, paste a pile into ChatGPT, pull the answer into Word, and hit send.
None of that is (or should be) the work.
That’s the cost or tax you pay before you get to the work. The hidden cost of moving information so you can actually decide, write, sell, hire, or help someone. You did not take the job to be a human copy machine. You took it to use your judgment, and most days the copy machine still feels like what you are doing.
We have confused moving the work with doing the work.
You leave the office, or you close the laptop at the kitchen table, and someone asks how the day went. You were busy. You can see the open tabs. You still cannot name the thing you actually did for a person. Busy is not the same as useful. The day asked you to carry work between rooms instead of doing the work and creating “real” value.
Gmail to Excel to the chat thread to the old PowerPoint to the search bar to ChatGPT to Word to send. Smart people, paid well, stitching tools together by hand so a meeting can start on time. By the time the meeting starts, half the morning is gone, and you have not used the judgment you were hired for.
There’s a word for the layer that’s supposed to do that moving. Harness.
What I Mean by Harness
A harness is what turns a smart model into a day you can actually use.
The model is the engine. The harness is the ride. You can park a very smart engine in the driveway and people will walk past it, because they do not care how impressive the engine is. They care about where they’re going, and whether the ride gets them there without making them do the work of the car. Nobody wakes up proud of an engine. They wake up needing to get somewhere.
Most of what we still call AI at work is an engine in the driveway. You walk up, you ask it something, it answers, and then you carry that answer back into the rest of your day and do the moving yourself. The intelligence was real but the experience is still you.
The best harness becomes invisible. You experience the outcome, not the orchestration.
We’ve spent a few years staring at the engine, arguing about how big it is, how fast it talks, how well it writes a paragraph and I think that is the wrong scoreboard. An engine sitting in the driveway doesn’t take anyone anywhere, and a weaker engine inside a real ride will beat a brilliant one you still have to push down the street.
A demo can make a room lean forward because the sentence on the screen is clean. Then the meeting ends and everyone goes back to the stitching described early. The engine performed but the ride never started.
You’ve Already Used a Harness
You don’t need me to explain the model. You’ve used a harness before. You just didn’t call it that.
Uber didn’t invent the car, the driver, or the road. Taxis had all three for a hundred years. What Uber built was the harness: it knew where you were, found the nearest driver, showed you the price before you agreed to it, and routed the car without you calling a dispatcher who might or might not pick up. The engine didn’t change but the ride did.
Waze didn’t invent the road either. It took the same map everyone already had and wrapped a harness around it. Live traffic, a reroute the second something ahead of you goes wrong. You stopped thinking about the map and started thinking about getting there.
TurboTax didn’t invent the tax code. The rules were always public, always available, and mostly unreadable. The harness was the interview. Answer a few plain questions, and the software pulls the right form, checks your math, and flags what’s missing, so you never have to read the code to follow it.
Each of those replaced a person doing the moving with a system built to do it for them. That’s what a harness does. It doesn’t add power. It removes the parts of the trip that were never the point.
Work hasn’t had its Uber moment yet. Right now, most AI at work is still the taxi standing on the corner. The engine is available if you know where to find it and how to ask. You are still the dispatcher, the router, and the tax preparer, digging through Gmail and Excel because nobody built the app.
That gap, between the taxi on the corner and the app in your pocket, is the harness gap I keep talking about. Most companies bought the car. Almost none of them built the app.
Ready for Your Judgment
Say it out loud: I need to prepare for tomorrow’s customer meeting.
A great chief of staff doesn’t start writing, and doesn’t open a blank page and hope a paragraph appears. They pull the last conversation, check which numbers don’t match what you’ll be asked about, find the slide you used six months ago so you don’t have to hunt for it, and put the risk in one place and the ask in another. They gather, they check, they organize and then they hand you something ready for your judgment. Yes, I said it. “Ready for your judgment”. That’s what should be the job.
You can feel that difference the second you sit down. One version of tomorrow has you hunting. The other has you choosing and practicing.
Early ChatGPT was a smart engine with a thin harness. You asked a question, it gave you an answer, and then you were the one who had to go find the email, check the number, open the old deck, and get it into someone’s hands. We taught ourselves a sentence that sounded advanced: answer my question. That sentence is too small now. The better sentence is help me accomplish this goal. The first keeps you in front of the engine. The second is what you say to a chief of staff and it takes steps off your hands so you can stay on the judgment. This is so important as we enter the world of agentic AI.
Do not measure your AI by how impressive its answer is. Measure it by how much unnecessary work disappears.
If the answer is beautiful and you still spent the morning copying, your harness failed. If the answer is ordinary and the meeting is already briefed, the harness did its job. I’d take the second one every time. The human is still the one who knows what good looks like in that room, who can hear what the customer or fellow leader didn’t say, and who decides. The harness exists so that person isn’t exhausted before the meeting or key decision moment even happens.
You Are the Harness Right Now
Look back at that day. Gmail to Excel to the chat thread, the old PowerPoint, ChatGPT, Word. That path is the proof. You are currently the harness.
I hear someone saying they already have and use AI, and I believe them. They have an engine. The true measure is how you tell whether they have a ride.
The model sat in the middle of the path and did one useful thing, and you did every other useful thing. You remembered where the number lived, decided which screenshot the room needed, found the slide, judged whether the draft was safe to send, and carried the work from one box to the next because none of the boxes would carry it for you.
That’s why the day feels hard even when the answers look good. The thinking is usually good it’s the stitching that is what wears you out and stitching is what a harness is supposed to do. When there’s no harness, the human becomes one.
People will tell you this is just how work is. I don’t buy that and I am changing that one customer and interaction at a time. Work is the meeting, the decision, the customer, the person you’re trying to help. The path you took to get there is the tax you pay. Some of that tax is real. A lot of it is leftover from tools that were never asked to ride together.
This Is the Employee Experience
We talk about employee experience like it’s a portal, an intranet, a survey, or a better benefits page. The real employee experience is the day. Did the day make them feel useful, or did it make them feel like glue.
When people are the harness, the experience is friction. They’re tired of being the integration. If you lead people, this is the day your team is having. A good harness is an employee experience decision before it’s anything else. It gives people their judgment back.
That’s not an adoption problem. It’s a harness gap.
I’ve heard the other story. We bought the tool, we ran the training, and people aren’t using it. That story makes the people the problem. Look at the day again. People are using it, in the middle of a path the tool doesn’t finish, and then we call the leftover work resistance. They adopted the engine. Nobody gave them a ride.
Do not train people to compensate for bad orchestration; fix the orchestration. If your people are still copying from Gmail to Excel to a chat window or uploading documents to a model, ask how much of the day still requires a person to be the moving part. If the answer is most of it, the work in front of you is a ride to build.
Why This Isn’t Just “Agentic AI” With a Softer Name
You’ll hear this framed as agentic AI, orchestration, or the agent layer, and every major vendor is currently racing to claim it. Fair pushback: isn’t “harness” just a friendlier word for what Salesforce, Microsoft, and a dozen well-funded startups are already selling?
No, and the difference matters. Most of what’s shipping under “agentic” right now is still engine-first. Faster reasoning, more tool calls, longer chains of steps the model can take on its own. I think that’s horsepower, not harness. It’s a bigger engine, marketed as a ride. That ride has not been intentionally designed with the passenger in mind.
A harness isn’t measured by how many steps the model can chain together autonomously. It’s measured by how much of your judgment it protects. An agent that books the meeting, writes the summary, and drafts the follow-up without ever showing you the risk it buried in the numbers hasn’t built you a harness. It’s built you a driver you didn’t ask for. The point was never to remove you from the seat. The point should have been to stop making you walk to the car.
That’s the test I’d apply to any tool claiming this territory: does it hand you something ready for your judgment, or does it quietly make the judgment and hand you the output? Those look identical in a demo but they are not the same product.
Get Me Ready for Tomorrow
That’s the sentence I want you to be able to say out loud. Not summarize this thread. Not draft me an email. Get me ready for tomorrow. Notice the sentence again: Get me ready for tomorrow. You say that. You don’t type it into a search box the way you’d look up a restaurant.
That’s not a small detail. It’s the whole argument, compressed.
A search box trains you to think in queries. A harness should be built to hear a goal. You don’t walk up to a chief of staff and type a keyword string. You catch them in the hallway, you call them from the car, and you talk to them the way you’d talk to someone who already knows the account. Voice forces that shift. You cannot comfortably say out loud the kind of stitched-together prompt you’d type into a chatbot. You can only say what you actually want.
That’s what voice does for a harness. It removes the last translation step, the one where a real thought gets flattened into careful text so a machine can parse it. An engine still sitting in the driveway doesn’t care how you talk to it. A harness does, because the entire point of a harness is to meet you where you already are, which for most people most of the time is talking, not typing.
I think voice will end up being the most honest interface a harness can have. It doesn’t let you hide behind a query. It makes you say the goal out loud, and a harness that can’t handle that isn’t finished yet.
The model can be very smart and still fail that sentence, because intelligence is not the same as a ride. A ride is what happens when the intelligence is pointed at a goal, the moving is done, and a person is still at the helm. Human at the helm is the point. If the harness starts making the call instead of preparing you to make it, you got a driver you didn’t ask for.
I still want the person in the seat, and I want that person less tired. I want the morning back, the meeting to start with a brief instead of a scavenger hunt, and the customer conversation to be about the customer instead of who forgot the slide. You already know what that would feel like, because you’ve had a person do it for you once. Maybe it was only once, and the day still changed. You became more available to the part of the work that was actually yours.
Did unnecessary work disappear, or did we just get a nicer answer sitting on top of the same day?
Get me ready for tomorrow.
Let’s move together from Now to Next.
P.S. If you want to talk about any of this in more detail, email me at jason@nowtonext.ai. I read every one, and I reply. We’re doing interactive intensives with organizations right now to build the future workforce experience. If that’s the work in front of you, say so in the note.
About Jason
Jason Averbook is the co-founder of Now to Next, an adjunct professor of business, and a globally recognized thought leader, advisor, and keynote speaker working at the intersection of AI, human potential, and the future of work. He spent the last few years as Senior Partner and Global Leader of Digital HR Strategy at Mercer, helping the world’s largest organizations reimagine how work gets done, not by implementing technology but by transforming the mindsets, skillsets, and cultures that have to come first.
Over the last two decades, Jason has advised hundreds of Fortune 1000 companies and their leaders, founded Knowledge Infusion and Leapgen, authored two books on the evolution of HR and workforce technology, and become a world renowned keynote speaker who has delivered hundreds of talks on the future of work. His work challenges leaders to stop treating digital transformation as an IT project and start treating it as a human strategy.
Through his Substack, Now to Next, Jason shares honest, provocative, and practical insights on what’s actually changing in the workplace, from generative AI to skills-based organizations to emotional fluency in leadership. His mission is simple: to help people and organizations move from noise to clarity, from fear to possibility, and from now to next.
You can reach him at jason@nowtonext.ai or connect on LinkedIn.
I appreciate the translatable in real life examples! Thanks Jason