The most expensive software you’ll ever use is the kind you built yourself.
That’s not a typo. I know it sounds backwards. Building is free now, right? You just open Cursor, or Loveable, or Bolt, describe what you want, and poof, working software. Or you spin up OpenClaw on your machine, install a few community skills, and suddenly you’ve got a personal JARVIS texting you confirmations from your couch.
The upfront cost is basically zero. And that’s exactly the problem.
Because what nobody is talking about, what the YouTube tutorials and viral Twitter threads conveniently skip, is the invoice that arrives later. The one that doesn’t come in dollars. It comes in your time, your attention, and your sanity.
I call it The Automation Tax.
Wait, Didn’t You Already Write About This?
If you’ve been following my work, you might remember a piece I wrote called Vibe Coding Is the New Sourdough Starter. In that post, I made the case that vibe coding would follow the same arc as the pandemic sourdough craze: a burst of excitement, the realization that maintaining it is a daily commitment, and eventual abandonment.
That thesis still holds. But since publishing it, something bigger has emerged.
The conversation has expanded beyond vibe-coded apps. We’re now in the era of DIY AI agents, tools like OpenClaw that don’t just build things for you, they do things for you. They run commands on your machine. They browse the web. They manage files. They send emails on your behalf.
And the same invisible tax applies, except now it compounds.
This is the part I didn’t cover last time. Consider this Part 2.
The Seduction of the Self-Hosted AI Agent
Let’s talk about OpenClaw specifically, because it’s the poster child for this movement.
OpenClaw (which has gone through more name changes than a witness protection participant: Clawdbot, then Moltbot, now OpenClaw) is an open-source AI agent that runs locally on your machine. It’s racked up over 68,000 GitHub stars. It can execute shell commands, browse the web, manage files, and communicate with you through messaging apps. People are calling it “the closest bridge to JARVIS yet.”
And honestly? The demos are incredible. Imagine texting your computer from a coffee shop: “Hey, summarize that PDF in my Downloads folder and email the highlights to my boss.” And it just… does it.
I get the appeal. I really do.
But here’s the question nobody asks after watching those demos: What happens on day 90?
The Three Costs Nobody Compares
When most people evaluate whether to vibe code an app or set up a DIY automation, they’re doing a simple comparison in their heads:
“I could pay $20/month for this SaaS tool… or I could just build it myself for free.”
That’s a two-variable equation. And it’s wrong. Because there are actually three costs you need to compare:
- The cost of building — Nearly zero with AI. This is the seductive part.
- The cost of buying — A predictable monthly subscription for an established solution.
- The cost of maintaining what you built — Unknown, unpredictable, and compounding over time.
That third cost is the one everyone leaves off the spreadsheet. And it’s the one that will eat you alive.
With AI-generated code, emerging data indicates that maintenance costs can balloon to 3x the cost of development. Analysts are predicting $1.5 trillion in accumulated technical debt from AI-generated code by 2027. Technical debt already consumes up to 40% of IT budgets, with 60–80% of that going purely toward maintenance.
Those aren’t abstract numbers for enterprise companies. That’s the same math that applies to your vibe-coded CRM, your OpenClaw automation stack, and the custom dashboard you built last weekend.
The OpenClaw Reality Check
Let me be clear about something: I’m not anti-OpenClaw. The technology is genuinely impressive and I think it represents a real glimpse of where personal AI agents are headed.
But there’s a massive gap between what OpenClaw can do and what the average business person should rely on it to do.
Here’s what the enthusiasts don’t emphasize enough:
The setup isn’t trivial. One reviewer described it bluntly: “challenging setup process, broken configs, and hours of debugging before I got my first agent to respond.” That’s not a criticism from a hater, that was someone who ultimately loved the tool. But “hours of debugging” is already a cost most business owners can’t afford.
The security surface area is enormous. OpenClaw has full shell access to your machine. It can run commands, browse the web, and access your files. Cisco published a piece titled “Personal AI Agents like OpenClaw Are a Security Nightmare.” Snyk warned that it’s “one prompt injection away from disaster.”
The legal exposure is real. Attorney Mitch Jackson wrote a must-read piece calling OpenClaw a “Legal Time Bomb”, detailing how giving an AI agent the ability to send emails, access files, and execute commands on your behalf creates liability scenarios most users have never considered. When your AI agent sends an email that misrepresents something, or accesses data it shouldn’t, or takes an action you didn’t explicitly authorize, you’re the one holding the match. Jackson’s breakdown of the legal implications should be required reading for anyone considering running an autonomous agent on their machine.
You might read those warnings and think, “That’s overblown. I’m just using it for simple stuff.”
Maybe. But every skill you install, every integration you connect, every automation you configure expands the trust boundary of what this agent can access. And you’re the only person responsible for securing all of it.
Community support has limits. When things break (and they will) your support system is GitHub issues and Reddit threads. One test found that getting answers to basic questions took anywhere from 8 hours to 3 days. That’s fine for a hobby project. It’s not fine when your business automation stops working on a Tuesday morning.
The Problem Isn’t One Automation. It’s Twenty.
Here’s where The Automation Tax really reveals itself.
One vibe-coded tool that breaks once a quarter? Annoying, but manageable. One OpenClaw skill that needs updating after a dependency change? No big deal.
But that’s not how it works in practice.
What actually happens is this: you build one thing and it works great. So you build another. Then another. Each one is simple. Each one takes “just an afternoon.” Each one individually seems trivial to maintain.
Then you wake up one morning with 15 custom automations, 8 OpenClaw skills, and 3 vibe-coded micro-apps, and something is broken. You’re not sure which one. The error message is unhelpful. The AI that built it generates a fix that breaks something else. And you’ve now spent your entire morning doing IT triage instead of running your business.
You just volunteered for a job you never applied to.
This is what I call compound fragility. Each new automation doesn’t just add risk linearly, it multiplies the surface area for failure. Every piece depends on external APIs, browser behaviors, code standards, and security patches that update on their own timeline, not yours.
And here’s the kicker: the skills that made you build these automations (prompting AI, describing what you want) are completely different from the skills required to maintain them (debugging, dependency management, security patching). You’re great at the first part. The second part is a career.
The Decision Framework You Actually Need
I don’t want to just tell you “don’t build things.” That would be dishonest and unhelpful. Sometimes building is the right call.
What you need is a framework for evaluating when building is worth it and when you’re signing up for a maintenance nightmare. Here’s what I use:
Ask two questions about any project before you start:
Question 1: How complex is the solution?
- Low complexity = a simple script, a single-purpose automation, a basic tool
- High complexity = multi-step workflows, data handling, user-facing applications, anything with integrations
Question 2: How often do the things it depends on change?
- Low change rate = the APIs, standards, and platforms it relies on are stable
- High change rate = dependent on third-party APIs, browser behavior, security patches, or other software that updates frequently
Now plot your project on this grid:
Low complexity + Low change rate: Go for it. Build it. The maintenance burden will be minimal. Example: a script that renames and organizes files on your desktop.
Low complexity + High change rate: Proceed with caution. It seems simple, but you’ll be fixing it regularly. Example: a web scraper that breaks every time the target site updates.
High complexity + Low change rate: Seriously ask yourself: does this product already exist? Your time building it has a real opportunity cost, even if the maintenance burden is manageable.
High complexity + High change rate: This is the danger zone. This is where most people are vibe coding full applications and OpenClaw-automating complex workflows. Every dependency is a ticking clock. Every upstream update is a fire drill.
Most of the flashy demos you see on social media? They live in that last quadrant. And nobody shows you what happens six months later.
What the Average Business Person Should Actually Do
Let me speak directly to you if you’re a solopreneur, a small business owner, a creator, a freelancer, someone who isn’t a developer and doesn’t want to become one.
Your time is your most valuable asset. Every hour you spend debugging a broken automation is an hour you didn’t spend on sales, content, client work, or strategy. The Automation Tax isn’t paid in money. It’s paid in the thing you can never get back.
Here’s my honest recommendation:
Use AI to enhance your workflow, not to replace your software stack. AI is incredible for drafting content, analyzing data, brainstorming, summarizing documents, and accelerating your thinking. That’s where the ROI is enormous and the maintenance burden is zero.
Pay for established solutions when they exist. If someone has already built the tool you need, and they have a team maintaining it, updating it, and securing it, that $20/month subscription is the best deal in business. You’re not paying for the software. You’re paying to not maintain the software.
Reserve vibe coding and DIY automation for genuinely unique problems. If no product exists for your specific workflow, if you need something truly custom, and if it falls in the low-complexity / low-change-rate quadrant, build it. That’s the sweet spot where AI-assisted building actually delivers on its promise.
Treat OpenClaw and similar tools as power-user territory. If you’re technically inclined, comfortable with the command line, and willing to invest ongoing time in maintenance and security, OpenClaw is a genuinely powerful tool. But if you’re a business owner who values your time and didn’t sign up for a side-quest career in DevOps, this isn’t a shortcut. It’s a detour. And a long one.
The Real Opportunity in AI
Here’s what I think gets lost in all the vibe coding hype: the biggest opportunity in AI isn’t building software.
It’s thinking better. It’s having a conversation with an AI that challenges your assumptions. It’s feeding it your data and getting insights you would have missed. It’s drafting a proposal in 10 minutes that would have taken 3 hours. It’s brainstorming 50 headlines and picking the best one.
None of that requires you to maintain anything. None of it breaks when Chrome updates. None of it leaves a WebSocket attack vector open on your local machine.
The Automation Tax is real. The question isn’t whether you can build it yourself. You absolutely can. The question is whether the ongoing cost of ownership is worth it when your actual job is something else entirely.
For most people, the answer is no.
The smartest thing you can do with AI in 2026 isn’t building more software. It’s knowing when not to.
