A local business owner did the website thing right. She paid attention to Google. For years she's been at or near the top of the first page for the search that matters most to her — the one where somebody in town types what she sells followed by "near me." The calls came from that spot at the top. She could more or less feel it working.

Sometime in the last year, the calls started thinning out. Not a cliff. A drift. She checks her ranking half-expecting to find she'd slipped, and she hasn't. She's still right where she was. Nothing she can see has changed, and yet fewer people are finding their way to her.

Nothing changed about her spot in line. What changed is that fewer people are looking at the line at all.

The click is disappearing

For most of the internet's life, search worked one way. You typed a question, you got a page of links, you clicked one. The whole game for a business was to be the link people clicked. Everything filed under "SEO" was really in service of that single moment: rank high enough that someone picks you.

That moment is going away, and the numbers aren't subtle about it. In early 2026, more than 68% of Google searches in the US ended without anyone clicking through to another website — up from about 60% two years earlier. Less than a third of searches now send a click to the open web at all. A big part of that is the AI answer Google writes at the top of the page: where one of those AI Overviews appears, which is now more than a fifth of searches, the links below it lose roughly half their clicks by most measures. People read the summary and they're done.

And that's just Google's own front page. There's a second front door that barely existed two years ago. Google's AI Mode — the version of search that's a conversation instead of a list — crossed a billion monthly users this spring. Plenty of people now skip the search box entirely and ask ChatGPT or Perplexity "who should I call for X in my area," then act on whatever name comes back.

Put it together and the picture is this: a customer used to land on your website and make up their mind there. Now, more and more, something drafts that decision before they ever reach you — and a lot of the time they never reach you at all, because the answer was good enough.

You can't optimize for a click that doesn't happen

This is the part that trips people up, because the old advice still sounds right. "Rank higher." "Get to the top of Google." But you can rank first and still lose, because first place on a page nobody clicks isn't worth what it used to be.

The question worth asking now isn't "where do I rank." It's "when somebody asks an assistant about a business like mine, does it know I exist, and does it get my facts right?" That's a genuinely different job. The old one was about beating other businesses to the top of a list. This one is about being the clearest, most trustworthy set of facts the machine can find when it goes to build an answer — because it's building that answer whether your facts are clean or not.

Here's the good news, and it's real: an AI isn't doing anything magic when it recommends a business. It's assembling an answer out of whatever it can find about you, and it strongly prefers businesses whose story is the same everywhere it looks. Most of what moves the needle is unglamorous, and most of it is yours to control.

What actually makes an AI confident about you

A few things matter more than the rest, and none of them are growth hacks.

Make your facts agree everywhere. Same business name, same address, same phone number, same hours — on your website, your Google Business Profile, and every directory you're listed in. This sounds trivial and it's the biggest one. If your hours say one thing in one place and something else in another, or you're "Bob's Plumbing" here and "Robert's Plumbing LLC" there, the machine gets less sure it's even looking at one business, and it reaches for a competitor it's more confident about.

Answer the real questions on your own pages. Assistants pull from pages that plainly answer the things people actually ask: do you do this specific thing, what does it roughly cost, do you serve my town. A page that opens with a straight answer beats one that opens with the year the company was founded. A lot of what makes those facts easy for a machine to read instead of guess at is structured data and clean markup under the hood — the same foundation a well-built site should have anyway.

Keep the outside world agreeing with you. AIs lean hard on corroboration — recent reviews, consistent listings, being mentioned around town. It's the difference between a business that says it's good and a business a dozen other sources back up as real and active.

None of that is a trick. It's making sure the thing that's increasingly answering on your behalf has its facts straight.

One honest caution

This is early and it's moving fast, and that's exactly the environment where someone will try to sell you a guaranteed "rank #1 in ChatGPT" package. Be skeptical of that the same way you should have been skeptical of guaranteed Google rankings ten years ago. Nobody controls what an assistant says about your business. What you can actually do is remove every reason for it to hesitate — every inconsistency, every missing fact, every page that makes it guess. You don't win by gaming the machine. You win by being the least ambiguous answer it can find.

How I work, briefly

I'll be straight that this overlaps with what I already do. When I build or rebuild a site, the structured data, the clean fast pages, the accurate markup — the stuff that used to just be "SEO" — is baked in, because it's now the same work that makes a business legible to an AI. I don't sell a separate "AI search" package, and I'd be wary of anyone who does. It isn't a bolt-on. It's what building a site properly already means in 2026.

If you're reading this

If you've noticed the calls thinning out while your ranking hasn't actually moved, or you just want to know whether an assistant can find you today and whether it gets your details right, that's a conversation I'd rather have early than late. It's usually a smaller fix than people expect.

Drop me a line.

— Nathan

That's not a few projects that didn't pan out. That's almost all of them.

The interesting question isn't whether the technology works. By every benchmark, it works better every month. The interesting question is why so much of what's being spent on it disappears without a trace. The answer is finally getting named, and the way the big AI labs are responding to it tells you most of what you need to know about where the actual value lives.

What the labs just did

In May, OpenAI, the company behind ChatGPT, announced a new $4 billion subsidiary they're calling the Deployment Company. Its entire job is to send engineers into customer organizations and help them actually put AI to work. Not sell more software. Send people. They bought a London firm called Tomoro the same day to get a hundred and fifty of those engineers on staff from day one.

Anthropic (the company behind Claude) and Google are hiring for the same role. The industry name for it is Forward Deployed Engineer. The plain version: a software builder who works embedded inside a customer's business, sits with their team, learns how that business actually operates, writes the code that runs inside their real systems, and stays around after launch to keep it working.

A year ago, this job barely existed outside of a couple of firms. As of this spring, it's the most fought-over role in AI.

Why this category suddenly exists

The reason this category exists is something the demo videos don't show.

A demo runs on a clean slate. The model gets a clear question, the data it needs is already prepared, the answer comes back in seconds, the screen looks impressive. Nothing about that environment looks anything like a real business.

A real business runs on the specific way you take orders. It runs on the spreadsheet your bookkeeper trusts more than the official system, the customer who gets a different price for reasons that go back to 2023, the exception you make for the supplier you've worked with for eight years, and a half-dozen other workarounds nobody ever wrote down. It runs on the customer who always calls instead of using the form. It runs on the email that always gets forwarded to the wrong inbox first.

None of that is in the model. The model doesn't know any of it. So the moment you try to use AI on real work inside a real business, you find that everything interesting about the project sits in the gap between what the model can do in the abstract and how your business actually operates.

That gap can't be bought. It can't be subscribed to. The only way to close it is for somebody who understands both the technology and your business to sit down and build the piece that fits. That's what the new job title is naming.

The thing nobody quite says

Here's the part the announcements don't spell out.

The models aren't the bottleneck anymore. They haven't been for a while. AI capability has been doubling every few months for almost two years, and the gap between what's possible and what businesses actually get out of it has stayed roughly the same size. If anything it's gotten wider, because the ceiling keeps rising and the work of closing the gap doesn't get any easier when you raise the ceiling.

That work is labor. It's not a product. You can't fix it by buying a better subscription, switching to a smarter model, or adding another tool to the mix. The only thing that closes the gap is somebody sitting down with you, understanding what you do, and building the specific piece that fits your specific business.

That's what billions of dollars are now lining up behind.

What this looks like at small-business scale

Here's where it gets interesting if you run a small business.

The big labs need armies of Forward Deployed Engineers because their customers are enormous. A Fortune 500 company has six layers of approval, an IT department, a compliance committee, a legacy system from 1998 that has to keep running, and twenty different teams that all have to agree on what the new tool should do. Closing the gap for them takes a team because the gap is huge.

A small business doesn't have any of that. You have an owner who knows the business in their head. You have a handful of tools that mostly work. You have one or two real problems where the right piece of software would meaningfully change how the week goes.

At that scale, you don't need a team. You need one person who shows up, learns how you actually do things, builds the piece that fits, and stays around to keep it working. That's the same shape of work the big labs are spending billions on. It's just sized for the actual problem.

In some ways you're better positioned for this than a large enterprise. There's no board to convince. There's no legacy system that has to keep running. A decision that takes a 200-person company six months of meetings, you can make in a conversation. The piece that takes them a team of fifty to build, you might need built once, by one person, and then maintained occasionally.

How I work, briefly

I should be direct here. This is the work I do at Wallman Solutions. One person, scoping and building and maintaining the same project, sized for small businesses.

It's not a novel concept. The big labs putting a name on the category and writing billion-dollar checks to scale it doesn't change what the work itself is. It's helpful that they're describing it out loud, because it makes it easier to explain why buying an off-the-shelf AI product and hoping it fits keeps not working out the way the demo suggested it would.

If you're reading this

If you've been watching the AI news and wondering what any of it actually means for a business your size, or you've already tried a tool that didn't quite stick, that's a conversation I'd rather have early than late.

Drop me a line.

— Nathan

A small business owner gets sold on an AI agent that's going to handle her lead follow-up. The setup goes well. Within a week the agent is sending replies to new inquiries inside of two or three minutes, which is faster than she's ever managed it herself. The dashboard shows green. Replies sent. Conversations started. By every visible metric, the thing is working.

About three weeks in, she starts noticing things. A referral she'd met for coffee got a generic cold-style pitch that landed weirdly. A lead who'd already booked got two follow-up messages asking if she wanted to book. A regular customer who'd asked a simple question about hours got something that read like a sales script.

She turns the agent off and goes back to handling it herself on Sunday nights. Slower, but at least she knew which conversations she was actually in.

AI is still a multiplier

I've written before about AI being a multiplier on whatever's already in the chair. That was in the context of hiring developers, but the same idea applies to your business, not just to the people writing code.

Right now is the loudest year I can remember for selling automation. About 82% of small business employers say they've invested in AI tools, and a lot of that investment is going into agents that promise to handle whole workflows: lead follow-up, scheduling, intake, light support. Some of those products are good. None of them fix what they're pointed at. They speed it up.

That's the trade nobody quite warns you about in the demo. The demo shows the polished version of a process. Your actual process is messier than that, and the agent is going to run on the actual process.

What "broken" looks like at small-business scale

When I say "broken process," I don't mean anything dramatic. I mean the normal stuff that builds up in any business that's been operating for more than a year or two.

It's the lead follow-up that nobody really owns. Some weeks the receptionist gets to it on Wednesday afternoon. Other weeks you handle it yourself on Sunday night. Sometimes a week goes by before anybody notices the inbox at all.

It's the intake form that lands in three different inboxes depending on which page on the site the customer filled it out from, and only one of those inboxes is checked daily. It's the standing knowledge that the back-office spreadsheet needs to be updated on Tuesdays, which lives in exactly one person's head and has never been written down anywhere.

It's the pricing exception you cut for one good customer because of a conversation in 2023, sitting unwritten between you and that customer and nobody else. It's the customer status that's a phone call to one person, an email to another, and a CRM note for a third, where the three sources don't agree.

These aren't problems exactly. They're seams. The business runs on top of them because there's a person in the middle who knows which of the three sources to trust on a given Tuesday. Slow, but coherent.

What happens when you automate around them

When you put automation on top of all that, the seams stop being soft.

The agent doesn't know that the lead who came in through the contact page is the same lead who already replied through the booking widget. So it sends a follow-up. The system doesn't know which of the three inboxes is authoritative, so it picks one and acts on what it finds there. The pricing exception isn't in the rules the agent was given, so the agent quotes the regular price and the customer notices.

None of that is the agent's fault. It's doing the job. The job just happened to include a lot of judgment calls that used to live in someone's head, and now they don't.

What makes this different from a slow process going wrong is the speed and the confidence. A human handling lead follow-up on Sunday night might make a mistake, but they'd usually catch it the next morning. An agent makes the same mistake at four in the afternoon, sends a polished message, marks the task complete in the dashboard, and moves on. By the time you find out, it's already happened thirty times.

The thing that was keeping the old process honest was often the bottleneck itself. The person in the middle was slow, but they were a soft check. Strip them out without replacing what they were doing, and you get more output, more confident output, and more wrong output, in roughly equal amounts.

What actually pays off

The work that pays off, in my experience, is almost always the work that happens before the automation.

It looks unglamorous. Sit down with the owner. Walk through what actually happens when a lead comes in, from the first touch to the booked appointment. Find the places where there are three inboxes, or one undocumented exception, or one piece of tribal knowledge that's keeping the whole thing standing. Decide which of those need to be cleaned up, which need to be written down, and which actually need a tool.

What usually comes out the other side is smaller than the original pitch suggested. Sometimes there's no agent at all, just a tightened-up form, a single inbox, and a one-page document that finally writes down the rules. Sometimes there's a small piece of custom automation that handles the specific cleaned-up workflow, and it's a smaller, narrower piece of work than what was originally on the table.

I'm not anti-automation. I lean on it heavily in my own work. The point isn't that the tools are bad. It's that the tools amplify, and amplification only works in your favor if the thing underneath is something you'd want amplified.

If you're reading this

If you're a few months into an automation that isn't paying off, or you're about to sign for an AI agent and you're not sure whether your process is in shape for it yet, that's a conversation I'd rather have early than late.

Drop me a line.

— Nathan

A small business hires a developer, or a small shop, to build a custom tool. Maybe it's a customer portal. Maybe it's a scheduling system that ties into the software they already use. The work moves fast. Three weeks in, the demo looks great. Everyone signs off, the invoice gets paid, the site goes live.

For about a month, nothing's wrong.

Then a customer calls because a form they filled out never showed up in anyone's inbox. The login page hangs on slow connections. A simple report takes forty seconds to run. Or, worse, a stranger emails to say they were able to view another customer's account without logging in, and would you mind fixing that before they go public with it.

You call the developer. They're three projects into someone else's work. They poke at it for a few hours and tell you it'll need to be "partially rewritten." That's the polite version of starting over.

AI is a multiplier on whoever's in the chair

I don't want to sound like the guy telling you AI is overhyped. I use these tools every day, and I wrote a fair bit about how big the productivity jump is in the last post. The speed gains are real. The way one engineer can cover ground that used to need a small team is real.

But the speed comes from amplifying whatever judgment is already at the keyboard. AI is great at writing the code you ask for. It's not the thing that decides whether the code you asked for is the right code. That part still belongs to the person driving.

A developer who already knows what production-grade software has to survive gets an enormous boost from these tools. They know what real users do to a system, what load does, what goes wrong at three in the morning. A developer who doesn't know any of that gets the same boost. Their output just looks done a lot sooner. The demo runs. The screenshots look right. Nothing is visibly wrong on the surface.

That's the problem. The demo doesn't tell you whether anyone in the room knows what they're looking at.

What "looks done" can hide

The kinds of trouble I'm describing aren't exotic. They're the same mistakes engineering teams have been making for decades. AI didn't invent them. It just made them faster to ship.

The first one is maintainability. Code that works the first time is not the same as code anyone can change later. I've watched builds go like this: the original developer gets the demo working, gets paid, walks away, and the first time the business needs a small change (a new field on the form, a different way of grouping customers) the next developer has to spend two days untangling what's there before they can write anything new. The cost of the second change ends up being higher than the cost of the original build. By the third or fourth change, the business is paying to rebuild the same feature twice.

The second one is stability under real load. A demo runs with one person clicking around. Production runs with twenty real customers using it at the same time, or with a database that's been growing every day for six months. Code that didn't think about either of those works fine right up until it doesn't. The kind of thing that's invisible during a one-week build and unmissable on a Tuesday afternoon when the page won't load and you have no idea who to call.

The third one is security, and it's the hardest one to spot from the outside, because the symptom is silence. A login that doesn't actually verify who you are. User-supplied data going straight into a database query in a way that lets a bad actor read whatever they want. Permissions checked only on the screen the user sees, not in the code behind it. You don't find out about any of this on launch day. You find out when somebody else does, and by then the conversation is about damage control, not features.

The bait-and-switch problem

There's a related thing worth knowing about the way custom software is sold. Some shops put a senior architect in the room to win the work, then hand the actual build to junior or contracted developers once the contract is signed. This isn't always shady. It's how a lot of larger agencies stay profitable. But the person who sold you on the project is not the same person making day-to-day judgment calls about your code, and that gap is where a lot of these problems live.

AI changes the dynamic in an uncomfortable way. The senior can still spot trouble in a code review if they're looking. The volume of code being produced has doubled or tripled, though, so there's more output to skim, less time to look closely, and a less-experienced developer using the same tools is now generating in an afternoon what used to take a week. The pace makes the gaps easier to miss, not harder.

If you ask a shop who's actually doing the work and you can't get a name out of them, that's information.

What to actually look for

You don't need to be technical to ask better questions during the sales process. A few that I'd want a friend asking before they signed:

Ask who specifically will be writing the code. Not "our team." A person, with a name, and ideally a sample of past work you can poke at. If you can talk to a previous client of theirs, do.

Ask what happens after launch. Does the same person stay on to maintain it, or do they hand you a zip file and disappear? When something breaks at 9 PM on a Saturday, who do you call?

Ask, in plain English, how they handle login security, what happens if your traffic doubles, and how they decide when something is ready to ship. A confident, plain-language answer to those questions tells you more than a slick deck does. If the answers are vague, or you get a wall of jargon designed to make you stop asking, that's also information.

The point isn't to catch anyone out. It's to find out whether the same person selling you on the work is actually doing it, and whether they've thought past the launch demo.

How I work, briefly

This is where I'll stop pretending I don't have a horse in this race.

Every project I take on at Wallman Solutions is scoped, built, hosted, and maintained by me. There's no handoff to a less-experienced team after the contract is signed. AI is a real tool that I lean on heavily, and it's a big part of why a one-person operation can do the work of a small team. Twenty years of writing software for production systems is what I lean on to keep that tooling honest. The combination is the whole point.

I'm not the right fit for every project. If you need a six-engineer team and a project manager, you don't need me. But if what you actually need is someone competent who picks up the phone, builds something that holds up under real use, and is still around in six months when the business has questions or a small change to make, that's the work I'm here for.

If you're reading this

If you're sitting on a prototype that looked great in a demo and is starting to wobble in production, or you're a few weeks away from signing a custom-software contract and you'd like a second set of eyes on the proposal before you do, that's a conversation I'd rather have before the rewrite than after.

Drop me a line.

— Nathan

For most of my career, when a friend or family member came to me with a software idea, my answer was the kind of soft "yeah, maybe someday" that everyone knew really meant no. A real custom build is a hundred hours of unglamorous work (auth, hosting, deploys, edge cases, the boring 80 percent), and a hundred free hours wasn't something I had to give. So the asks piled up. I wanted to help every time, but the math wasn't there, and nothing actually got built.

That started to change in early February, when the latest round of models hit. Two pretty different things shifted at once.

I needed somewhere to actually use this stuff

The pace of AI right now is the fastest I've ever watched a technology move, and I've been doing this long enough to have seen a few inflection points. New models, new tools, new agentic patterns. The ground under what's possible keeps shifting week to week. You can read about all of it and still not really know how it works. The only way I've ever learned a new technology is by actually building something with it.

The catch is that my day job, like a lot of jobs in regulated or risk-averse industries, has real limits on where and how AI can be used. There are good reasons for those limits and I'm not complaining about them. But it does mean the slice of AI I get to touch professionally is small compared to what's actually out there. If the rest of my exposure comes from reading what other people are writing about it, I'll always be a year behind whoever's actually shipping.

So I needed somewhere of my own. A place where I could take an idea all the way through: pick the model, wire it up, deploy it, watch it break, fix it, do it again. Side projects do that better than anything else I've found. A weekend on something real teaches you more than a stack of conference talks.

And the math finally adds up

For my whole career, the productivity ceiling on one engineer working alone was basically fixed. You got a little faster with experience, a little faster with better tooling, a little faster with a focused weekend, but the curve was gentle. The hours a real custom build needed cost more than I had to give.

AI tools have started to break that ceiling, and not by a small amount. GitHub's research on Copilot a couple of years ago put it at about 55% faster, and the current generation of models is in a different category. If you've read studies suggesting AI tools don't help much, or even slow developers down, you're reading studies that measured last year's tools. The jump that happened in early 2026 is real, and on the kind of work I'm actually doing on Saturdays it's much bigger than even those older numbers suggest.

New projects from scratch, full-stack, solo, with my tooling dialed in and no legacy code in the way. That's where these tools are at their best. I'm a heavy user of them and I've put a lot of unpaid hours into learning to drive them. The multiplier I'm seeing is 3-4x, sometimes more on a project that fits the tools well. A weekend now does what a two-week sprint used to.

AI absolutely makes you faster. But it's a multiplier, not a substitute, and it amplifies whatever you bring to it. A 3-4x boost on top of twenty years of production engineering experience looks like real software shipped in a weekend. The same boost on a developer who doesn't already know what to build, how to structure it, or how to keep the model on the rails just produces bad code faster. The tools are remarkable. They still need someone in the chair who knows what they're looking at.

So the project that used to take me a hundred hours takes about twenty-five. A friend asking for a small custom tool isn't a six-month favor I have to politely decline anymore. It's a real deliverable on a real timeline at a price that makes sense for both of us.

Why a company, and not just a hobby

I thought for a while about just keeping this as favors and never putting a name on any of it. Two things eventually changed my mind.

One was that the asks kept coming. Friends tell their friends, and "I know a guy" travels fast in small business circles. Without a real name and process behind the work, every project ends up being its own one-off scramble. With one, I can take on more without losing track of things in my inbox.

The other was that writing the code itself is rarely the hard part. What people actually need is code that's scoped fairly, hosted somewhere it isn't going to disappear, and maintained by someone they can actually call. That last part is rarer than people expect. Plenty of freelancers will build you a thing and walk away. The site goes live, the contract closes, and six months later when the database fills up you're calling someone new who has to start from scratch.

Wallman Solutions is built to be the opposite of that. Every project I take on is built, hosted, and maintained by me, the same person who scoped it. AI is what makes it possible for one engineer to cover that much ground. But the reason it's worth doing is that most small businesses don't actually need a small team. They need one competent person who picks up the phone.

If you're reading this

If you're someone I've told "yeah, eventually" to in the last few years, the answer's different now. Send me what you were thinking about. There's a good chance it's something I can actually build.

And if you're running a small business with a process held together by spreadsheets, or you've got an AI-generated prototype that looked great in a demo and is starting to fall apart in production, that's the same kind of work and I'd love to hear about it.

Drop me a line.

— Nathan