AI Didn't Replace Our Workers. It Replaced Our SaaS Stack.

May 14, 2026·9 min read

Time magazine ran a piece this week on small businesses replacing workers with AI. They mentioned my company. Here's the version of the story that's more useful to other operators.

The Time article, published May 14, 2026.

In November 2025, Claude Opus 4.5 came out, and I realized I could rebuild a billion-dollar company's software stack with a small team and a few months of focused work.

I run Sonora, a premium online guitar mentorship company with over 6,000 graduates and a team of around 30. We've been operating for seven years. For most of that history, we did what every other modern small business does: we paid HubSpot, Calendly, Vimeo, DocuSign, and a dozen other SaaS vendors for the operational backbone of the company.

By April 2026, all of that was gone. We'd replaced our entire SaaS stack with internal tools built on Supabase, Vercel and Claude. The cost difference is about $250,000 per year. The capability difference is bigger than the cost difference.

That's the actual story. The Time piece focused on the headcount changes that followed, which is fair journalism, but it's a downstream effect of the real shift. We didn't decide to cut headcount and then automate around it. We rebuilt the operational layer, and a portion of the work it used to coordinate stopped existing. The headcount math was second-order.

In parallel, I launched a class called Pioneer Species to upskill everyone on my team plus around 80 non-technical friends into AI engineering. If there's a single word for what this story is actually about, it isn't displacement. It's closer to upskilling.

Get the framing right and the rest of the playbook is obvious. Get it wrong and you'll spend the next 18 months running the wrong experiments.

What we actually rebuilt

The SaaS stack I inherited as the company scaled looked like every other modern small business:

HubSpot for CRM, marketing automation, and email
Calendly/Oncehub for scheduling
Vimeo for video hosting
Slack for student communication
Aloware for calls and SMS
DocuSign for contracts
A dozen other point tools for finance, ops, and support

Each tool was useful. Each tool was also a closed system. None of them talked to each other without an integration layer (Zapier, Make, n8n) that broke regularly. None of them could be customized to the specific workflow of an online guitar school. All of them charged per-seat fees that scaled linearly with the company.

When Claude Opus 4.5 launched, the equation changed. I could build internal versions of every one of those tools in weeks, not years. Not better than the SaaS versions on every dimension. But specifically shaped to our workflows, fully integrated by default, and owned outright.

The stack we have today:

Supabase as the single source of truth for student, sales, and operational data
Custom CRM built on Supabase, replacing HubSpot
Custom scheduling, replacing Calendly
Custom video hosting and progress tracking, replacing Vimeo
Custom split testing and analytics stack
Stripe as the only third-party we kept for payments
Custom messaging platform for student communication
Claude API as the intelligence layer that does the work agents used to be too limited to handle

This isn't a stack you could have built six months ago. Without Claude 4.5-class models, the agents couldn't reliably handle the long-running operational tasks the SaaS tools coordinated. With them, they can.

The agent layer wraps everything

Once your data lives in one place, agents become trivial to point at it.

This is the half of the story most operators miss. Replacing the SaaS stack is the architectural shift. The agent layer on top is what makes that shift compound.

At Sonora today, anyone on the team can ask Claude for arbitrarily complex things about the business and get an answer in seconds. "Which mentors had the strongest student outcomes last quarter, broken down by genre focus?" "Show me every student who's stalled out in the last 30 days and what's going on with them." "Draft re-engagement messages for the 14 students who haven't logged in this week, personalized to where they are in the curriculum." A year ago each of those was a small project. Now they're prompts.

The same is true for actions, not just insights. Agents schedule and reschedule sessions, draft outbound messages, flag churn risk, kick off onboarding sequences, reconcile billing edge cases, and run multi-step workflows that used to require a human stitching three tools together. When the system underneath is unified, an agent can operate across the entire surface area of the business without anyone having to wire up ten APIs and a pile of Zapier glue.

This is where the framing matters most. Yes, we eliminated some roles. But the work that remains is dramatically higher-leverage than it used to be. A single operations person at Sonora has the equivalent of a small department at their fingertips. A mentor can pull deep context on a student in seconds before a session instead of spending fifteen minutes hunting through three different tools. Our sales team can run analyses today that would have required hiring a dedicated analyst a year ago.

The honest version of this story is that the consolidation didn't only eliminate jobs. It elevated the remaining ones. People at Sonora are doing work that would have been senior-leadership work two years ago, because the agent layer collapses the distance between wanting to know something and knowing it, or wanting to do something and having it done.

The press default of "fewer people" misses the actual shift. It's the same team, slightly smaller for now, doing different work on top of a substrate that supercharges each person's capacity. That doesn't mean we'll need fewer people long-term. It means each individual's impact scales dramatically, and we're not pausing hiring. We may well hire more from here. The difference is that future hires will be oriented around people who already have an AI-first skillset.

What stopped existing once we rebuilt

When you replace your entire software layer with a unified internal system that has Claude in the loop, a lot of administrative work stops being necessary. Not because AI is doing that work in a different way, but because the work itself was an artifact of the SaaS architecture.

Examples:

Manual data reconciliation between HubSpot and Stripe. Gone, because the data lives in one place now.
VA work coordinating between Calendly and our internal scheduling rules. Gone, because scheduling is built around our rules natively.
Onboarding admin: setting up accounts, sending welcome sequences, scheduling first sessions. Gone, because the rebuilt system does this automatically when a student signs up.
Manual outreach for cold leads. Reduced significantly, because our sales team's workflow was rebuilt around AI-assisted prospecting.

The headcount changes the Time piece described followed from this. Most of the work we eliminated was contractor and VA work, not full-time roles. The setter team contracted from 12 people to a much smaller number. We also moved off a dedicated sales manager to a player-coach model, because the front-line-to-executive translation that role used to provide was now basically immediate. Our main operations person is still on the team.

People lost income, and some of those people had been with us for years. That's a real cost of the transition and worth acknowledging directly rather than burying under productivity metrics.

But the work that was eliminated was work the SaaS architecture created. Not work that produced student outcomes. The distinction matters because it tells you where to look in your own business.

What we kept human, and why

The Time piece used a single number, 30 employees, and didn't break it down. Here's the breakdown that matters:

18 of those 30 people work directly with students every day. They're mentors and student success staff. They are the product. Sonora is not a SaaS company. We sell mentorship from professional guitarists to serious students. The human relationship is the entire thing.

The automation investment didn't reduce this team. It protected and grew it. The operational overhead the SaaS stack required was eating real management attention, and most of what we freed up went into expanding what mentors can do for students.

This is the framework I'd offer to other operators: find the line between the work that produces your actual product and the work that exists to coordinate the software running the company. Automate ruthlessly on one side of that line. Invest aggressively on the other.

Most companies have this backward. They automate customer-facing work because it's expensive and visible, and keep the operational layer because it's familiar. The operational layer is what AI is best at. The customer relationship is what humans are best at, especially anywhere trust, taste, or judgment is the actual product.

Why everyone on the team learned to build

This is the part of the story I care about most, so it deserves more than the paragraph I gave it above.

In January 2026, I launched Pioneer Species to help non-technical people upskill into AI engineering and stay resilient through what's coming. I invited everyone on the Sonora team plus around 80 friends from wildly different fields, including things like music, medicine, law, film, and finance. None of them came in technical. By the end of the program, every one of them could spec, prompt, and ship a working internal tool. The non-technical people in my life had become developers. In the world we're heading into, that's not a nice-to-have. That's literacy.

We've run two cohorts so far and are moving to a recurring schedule. It's a single program where non-technical friends, business owners, and professionals across every field learn to build agentic systems together.

The point isn't to turn everyone into a software engineer. It's to give non-technical people enough fluency with AI tools that they stop being downstream of someone else's build queue and become builders themselves. When the person who knows the workflow can also build the workflow, the company gets faster, the work gets better, and the people doing it have more leverage than they've ever had.

I genuinely believe this is the single most important skill any working professional can learn over the next few years, regardless of what they do. The course is built for non-technical people on purpose. The shape of every job is about to change, and the people who learn to direct AI systems instead of compete with them will do dramatically better than the people who don't.

I want to spend my working life collaborating with as many talented people as possible, and I plan to keep doing it for decades. But the technology will keep getting more capable whether any of us are ready or not. The most useful thing I can do for the people I work with, and the people I care about, is help them get ahead of it instead of defending a version of work the world is leaving behind. Pioneer Species exists because I think that's the only humane option.

What I got wrong

A few things I'd do differently:

I moved too fast on the setter team. The economics were clear, but the transition was rougher than it needed to be. I should have run a longer parallel period, given affected contractors more runway, and been more explicit with the broader team about what the changes meant and didn't mean. The human cost of moving fast is real and I underweighted it.

I undercommunicated to the team. When you're rebuilding the stack, your team watches in real time and draws their own conclusions about who's next. I should have been clearer earlier about which roles were and weren't in scope.

I overestimated my own clarity on the framing. Speaking to a reporter on the record, I described what we did in language that read as "we replaced workers with AI." That's not what I think happened. But it's what I said. The lesson is that the framing has to be sharp in your own head before you take it public, because the version you say off-the-cuff is the version that gets quoted.

What this means for other operators

If you run a small or mid-sized business and you've been waiting for the right time to make a serious AI push, the right time was six months ago. The second-right time is now.

The specific lever isn't "use AI to do your team's work." It's "use AI to collapse your SaaS stack into something you own, and let the operational simplification do the headcount math from there."

A few principles I'd offer:

Don't start with the headcount question. Start with the software architecture question. Headcount changes are downstream of architecture changes.
Keep humans where humans matter. In most businesses, that's the customer relationship. In service businesses, almost always. Automate aggressively everywhere else.
Build the system before you cut the team. Run parallel periods. Confirm the AI version actually works on real workflows for real customers before you make people redundant. The cost of being wrong is high in both directions.
Communicate constantly with your team during the transition. They are watching. Their assumptions in the absence of information will be worse than reality.
Upskill the team. Give every person enough AI fluency to become a builder, not just a user of the system you put under them. A team where each person can spec, prompt, and ship is dramatically more capable than one that can't.

What I'm working on next

Pioneer Species is where I teach all of this, and it's built for non-technical people in any industry. The class covers AI engineering fundamentals, harness engineering, multi-agent systems, and building the entire agentic layer for a company. The Sonora rebuild runs through it as a worked example: the architecture, the tools, the specific prompts and workflows, what worked, what didn't, what I'd do differently. The name comes from the first species to breathe life into an ecosystem after a major disruption, which is what the early phase of AI-native company building looks like to me right now.

The Time piece will probably outlast this one in search results. But this is the version I'd want other operators to read.