DEV Community: Augmented Mike

I Automate Businesses for a Living. Here's What Actually Works and What's a Complete Waste of Money.

Augmented Mike — Wed, 15 Apr 2026 13:07:39 +0000

Every software vendor on the planet has slapped "AI-powered" onto their product page in the last two years. Your inbox is full of pitches. Your LinkedIn feed is full of people claiming AI changed their life. And you're sitting there running a real business, wondering which of this stuff is actually worth your time.

I've been building software for 25+ years. I've built production AI systems — not demos, not prototypes, real things that run in production and handle real money. I built ClaimHawk, which automates dental insurance claims processing and cut denials by 67%. I've helped small teams replace manual workflows with AI-powered automation that actually saved them time and money.

Here's what I've learned about AI and small business: most of it is noise, some of it is transformative, and the difference comes down to whether you're solving a real problem or buying a solution looking for one.

The honest truth about AI for small business in 2026

Let me get this out of the way: AI is not going to run your business for you. If someone tells you that, they're selling something that will ultiimately disspoint you.

What AI can do is handle the repetitive, tedious, time-consuming tasks that eat your day. The stuff you hate doing. The stuff you keep meaning to hire someone for but can't justify the salary. Data entry. Email sorting. Invoice processing. Report generation. Customer question routing. Content drafts. Document formatting. And now, a bunch of "omputer use" tasks.

That's where AI earns its keep. Not in some grand "digital transformation" — in the boring, practical, everyday grind that steals hours from the work that actually grows your business.

What's actually working right now

I'm going to be specific here because vague advice is worthless. Every "AI for business" article gives you the same generic list — chatbots, content generation, analytics. None of that tells you what to actually do on Monday morning.

What follows are categories of AI automation that I've either built for clients or seen work consistently in small business contexts. Not theoretical use cases pulled from a vendor whitepaper. Actual systems running in actual businesses, handling actual money. I've included what the problem looks like before automation, what the solution does, and roughly what it costs to build — because "it depends" is not an answer.

Document processing

If your business touches paper — invoices, contracts, applications, insurance forms, receipts — AI can read those documents, extract the data, and put it where it needs to go. Not "eventually" or "with some training." Now. Today. The OCR and language understanding models are good enough that they handle 90%+ of standard business documents without human review.

I built exactly this for dental practices with ClaimHawk. Insurance EOBs come in as PDFs or scans. The system reads them, extracts denial codes, cross-references with patient records, and either processes the claim or flags it for human review. Before this existed, someone was doing that by hand for 15-20 hours a week.

The pattern applies to any business drowning in documents. Law firms processing discovery. Real estate companies handling applications. Accounting firms sorting receipts. If someone on your team is copying data from documents into a system, that's automatable.

Customer communication routing

This isn't about building a chatbot. Chatbots are usually terrible and your customers hate them. This is about using AI to sort, prioritize, and route incoming communications so the right person sees the right message at the right time.

A support email comes in. AI reads it, determines whether it's a billing question, a technical issue, or a sales inquiry. It routes to the right person. It drafts a response that the human can edit and send. The human still handles the relationship. The AI handles the triage.

This works for email, support tickets, contact form submissions, even voicemail transcripts. The key is that the AI isn't replacing the human interaction — it's eliminating the sorting and routing time.

Report generation

You know that monthly report you spend three hours on every month? The one where you pull data from your CRM, your accounting software, and your project management tool, paste it into a spreadsheet, make some charts, and email it to your partners?

AI can do that. Not "AI" as in some $500/month SaaS platform. A custom script that connects to your existing tools, pulls the data, formats the report, and delivers it to your inbox on the first of every month. Total cost to build: 10-15 hours of engineering time. Total cost to run: near zero.

This is one of the highest-ROI automations I build for small businesses. It's not glamorous. Nobody posts about it on LinkedIn. But it saves hours every month, eliminates errors, and the report is always on time.

Internal knowledge bases

Your business has institutional knowledge trapped in people's heads. When Sarah knows how to handle the weird edge case with the Johnson account, and Sarah goes on vacation, nobody knows what to do.

AI-powered knowledge bases solve this. Not a wiki that nobody updates — a system that ingests your existing documents, emails, Slack messages, and SOPs, and makes them searchable in plain English. "How do we handle refunds for the enterprise plan?" gets a real answer pulled from the actual documented process.

The technology for this (called RAG — retrieval augmented generation) is mature and reliable. The hard part isn't the AI. It's getting the business to document its processes in the first place. But once the documentation exists, even imperfectly, the AI makes it actually useful.

What's not working (despite what the vendors tell you)

AI chatbots for customer service

I said it above but it deserves its own section. The vast majority of AI chatbots deployed by small businesses make the customer experience worse, not better. They're trained on your FAQ page, which doesn't cover the actual question the customer has. They hallucinate confidently wrong answers. They frustrate people who wanted to talk to a human.

There are exceptions — if you have a very narrow, well-defined set of questions (like "what are your hours" or "how do I reset my password"), a chatbot can handle those. But if your customer questions involve any nuance, context, or judgment, a chatbot is going to hurt more than it helps.

The better approach: AI-assisted human support. The AI drafts responses, suggests relevant documentation, and summarizes the customer's history. The human reads, edits, and sends. You get the speed benefit without the customer frustration.

"AI-powered" versions of tools you already use

Your CRM added an AI feature. Your project management tool has AI now. Your email client has AI compose. Are any of these worth using?

Honestly, most of them are mediocre. They're built by teams that specialize in CRM or project management, not AI. The AI features are bolted on to justify a price increase or a press release. They work well enough for simple tasks — summarizing a long email thread, suggesting a meeting time — but they rarely do anything you couldn't do yourself in 30 seconds.

The exception is coding assistants (GitHub Copilot, Cursor, Claude) which have genuinely changed how software gets built. But those are relevant if you're a developer. For most small business owners, the AI features in existing tools are nice-to-haves, not game-changers.

Fully autonomous AI agents running your business

The dream: AI that handles your entire sales pipeline, manages your projects, responds to customers, and makes strategic decisions. The reality: we're not there yet, and anyone selling you this is selling fantasy.

Autonomous agents are powerful in narrow, well-defined domains (like processing insurance claims in a specific format). They fall apart when they need judgment, context, or the ability to handle situations they haven't seen before. For a small business with a million edge cases and human relationships at the center, fully autonomous AI is a recipe for disaster.

The right approach: human-in-the-loop. AI handles the repetitive parts. Humans handle the judgment calls. The boundary between those two shifts over time as you build trust in the system. But starting fully autonomous is asking for trouble.

How to figure out what to automate in your business

Don't start with "what AI can do." Start with what's eating your time.

Write down every task you did last week. Every one. Now sort them by two criteria: how much time did it take, and how much judgment did it require?

The sweet spot for AI automation is high time, low judgment. Data entry. Formatting. Routing. Sorting. Copying data from one system to another. Generating reports from existing data. These are the tasks where AI delivers immediate, measurable ROI.

The danger zone is low time, high judgment. Strategic decisions. Relationship management. Creative work. Negotiation. These are tasks where AI can assist (by gathering information, drafting initial versions, summarizing context) but should not decide.

The tasks in the middle — moderate time, moderate judgment — are where you need a conversation with someone who understands both your business and the technology. That's what I do.

The build-vs-buy decision

For most small businesses, the right answer is a mix of both.

Buy off-the-shelf tools for generic problems. Email. Calendar. CRM. Accounting. These are solved problems and the existing tools are good enough. Don't build a custom CRM unless your workflow is genuinely unique.

Build custom for problems specific to your business. The weird workflow that no SaaS tool handles. The report that requires data from three different systems. The document processing pipeline that's specific to your industry. The integration between tools that don't natively talk to each other.

The mistake most small businesses make is trying to force generic tools to handle specific problems. They end up with five Zapier automations, three spreadsheets, and a process that breaks every time someone changes a field name. That's more expensive than building the right tool in the first place.

What it costs

This is the part everyone wants to know and nobody wants to answer directly. So I'll answer it.

My rate is $150/hr. Most small business automation projects take 20-60 hours. That's $3,000-$9,000 for a custom tool that solves a specific problem in your business.

For context: a junior developer from an offshore agency will charge you $30-50/hr but take 3-5x as long, with multiple rounds of rework. Total cost ends up about the same, but it takes three months instead of two weeks and you spend half that time managing the project.

An established US agency will charge $200-400/hr. Same work, higher overhead, project managers and account reps between you and the person writing the code.

A SaaS tool that does roughly what you need costs $50-500/month forever. Over two years, that's $1,200-$12,000 — and you don't own anything. You're renting a tool that does 70% of what you need and forces you to work around the other 30%.

The custom build costs more upfront and pays for itself within months. You own it. It does exactly what you need. It doesn't have features you don't use or missing features you need. And there's no monthly bill to keep your own software running.

Where to start

If you've read this far and you're thinking "okay, I have some tasks that fit the high-time-low-judgment pattern," here's what to do:

First, pick one process. Not three. Not the whole business. One process that takes too much time and doesn't require much judgment. The monthly report. The invoice processing. The customer email routing. One thing.

Second, document what happens today. Write down each step, how long it takes, how often it happens, and where the data comes from and goes to. This doesn't have to be fancy. A bulleted list is fine.

Third, talk to someone who builds this stuff. Not a vendor who's selling you a platform. Not a consultant who's going to spend three months on a "digital transformation roadmap." Someone who will look at your process, tell you what's automatable and what's not, and give you an honest estimate.

That's what I do. A 30-minute call is enough to figure out whether your problem is worth automating and what it would take. If it's not worth it, I'll tell you that and save you the money.

The bottom line

AI automation for small business is real, it works, and it's accessible at a price point that makes sense. But it's not magic, it's not autonomous, and it's not going to replace your judgment or your relationships.

The businesses that benefit most from AI are the ones that approach it practically: identify a specific problem, build a specific solution, measure the results, and expand from there. Not the ones chasing the latest tool or trying to "transform" everything at once.

ML/AI Development & Fractional Services | github.com/augmentedmike | youtube.com/@augmentedmike | x.com/_augmentedmike

I Built an AI Agent That Writes All My Production Code. Here's What I Learned.

Augmented Mike — Tue, 14 Apr 2026 18:14:00 +0000

I'm Mike. I've been writing software for 25+ years — systems programming, web, mobile, cloud, and now AI. I'm currently building two things that I think this community would find interesting, so I figured I'd introduce myself and share what I've learned.

AM — my autonomous coding agent

AM is an AI agent that writes all of my production code. Not copilot-style autocomplete — fully autonomous task execution. I give it a ticket, it reads the codebase, writes the code, runs the tests, commits, and moves on to the next ticket.

It built my entire portfolio site. Every page, every component, every deployment. I direct strategy and make architecture decisions. AM executes.

helloam.bot

The interesting engineering behind it:

Stateless by design. AM carries no memory between invocations. Every run is one-shot: read the state from files, do one unit of work, write the state back, exit. All state lives in markdown files and git history — todo.md, criteria.md, iteration logs. This sounds like a limitation but it's actually the key to reliability. There's no context window drift, no accumulated hallucinations, no state corruption. If a run fails, you just run it again. The files are the source of truth.

Three-tier memory. Short-term memory is markdown rules that get injected into every session — things like "never use deprecated API X" or "the database schema changed, use the new column name." Long-term memory is a SQLite FTS5 database with ranked search — lessons learned across projects. Episodic memory is git history and iteration logs. The system is modeled loosely on how human memory works: working memory, declarative memory, and episodic recall.

Gated state machine. Tasks move through backlog → in-progress → in-review → shipped with verification gates at each transition. The gates are enforced by code, not self-reported by the agent. "In-review" means every acceptance criterion has been verified against the actual output. The agent can't advance a task by saying "I think this works" — it has to prove it.

Worktree isolation. Each task gets its own git worktree. Multiple agents can run simultaneously on different tasks without stepping on each other. When a task ships, the worktree gets squash-merged into the integration branch. Clean linear history, no merge conflicts between agents.

The whole system is open-source. You can see it at helloam.bot.

ClaimHawk — AI automation with vision and action models

The other thing I built is ClaimHawk, which automates dental insurance claims processing. This one pushed me into territory that most AI projects don't touch: vision models and action models working together in a production pipeline.

Here's the problem: dental practices lose tens of thousands of dollars a year to denied insurance claims. Not because the work wasn't done — because the claim was submitted with the wrong code, a missing attachment, or a formatting error. Staff spend hours every week doing this manually, and they make mistakes because they're processing hundreds of claims.

The OCR challenge. Insurance documents (EOBs — Explanation of Benefits) are a mess. Multi-column layouts, inconsistent fonts, degraded scans, tables mixed with free text. Generic OCR engines like Tesseract choke on them. I'm using ChandraOCR, which handles the layout complexity that dental insurance documents throw at you. It runs locally — no document data leaves the network, which matters for HIPAA.

The vision/action model stack. Here's the part that surprised me the most. Insurance carrier portals don't have APIs. They have websites built in 2008 with session timeouts and CAPTCHA gates. ClaimHawk uses computer vision to navigate these portals the way a human would — reading the screen, clicking buttons, filling forms, uploading attachments. When a portal redesigns its UI (which happens constantly), the vision model adapts because it's reading the interface semantically, not relying on CSS selectors that break every time someone changes a class name.

Local models, not cloud APIs. ClaimHawk runs on fine-tuned Qwen3 models, not GPT or Claude. Patient health data can't leave the building under HIPAA. Open-weight models trained with RLHF on real dental claim data run on hardware the practice controls. The models understand dental terminology, CDT coding, and carrier-specific appeal formats because they were trained on thousands of real claims.

The results so far: 67% fewer denials, 4x faster payment cycles.

What I've learned that might be useful to you

If you're building AI systems that need to interact with the real world (not just generate text), here are the lessons that cost me the most time:

Vision models are more resilient than web scraping. I fought with Playwright and CSS selectors for months before switching to computer vision for portal navigation. The vision approach handles UI changes that would break any selector-based scraper. The initial investment is higher, but the maintenance cost drops to near zero.

Statelessness is a feature. Every agent framework I evaluated tried to maintain state in the model's context window. This creates debugging nightmares, context window limits, and accumulated errors. Making the agent stateless and putting all state in files made everything simpler — auditing, recovery, parallelism, all of it.

Local models are production-ready. The assumption that you need GPT-4 or Claude for everything is wrong. Fine-tuned open-weight models outperform general-purpose frontier models on domain-specific tasks, cost a fraction as much to run, and give you complete control over your data. The fine-tuning pipeline is the investment, but once you have it, you own the capability.

I'll be posting more about both of these projects — the engineering decisions, the mistakes, and the stuff that surprised me. Happy to answer questions about any of it.

ML/AI Development & Fractional Services | github.com/augmentedmike | youtube.com/@augmentedmike | x.com/_augmentedmike