The SMB Reality Check: What Agents Can & Can't Do
Stop trying to automate everything. Here are three wins you can ship today, and three traps that will quietly burn your budget.
Omri Dan · Nomadan founder
Most small business owners are getting two flavors of bad advice right now.
The first is "agents will replace your whole team." They won't. The second is "agents are still science fiction." They aren't.
The truth lives in a narrow practical band, and that band is where Nomadan operates. Here's what's actually working, and what's quietly draining bank accounts.
Three wins you can ship this quarter
1. Inbound triage. Email, support tickets, missed-call follow-ups. An agent that reads, classifies, drafts a reply, and waits for you to hit send is a 10-hour-a-week win for almost any service business. It does not need to be magical. It needs to be consistent.
2. Internal knowledge retrieval. Your team is asking the same five questions in Slack every week. Point an agent at your docs, your wiki, your past invoices. Let it answer. The ROI is measured in interruptions you don't have anymore.
3. Quote and proposal drafting. Structured input, structured output, a human reviewer at the end. This is the boring agent that pays for itself in week one.
Three traps that will burn your budget
1. Full autonomous "AI employee" pitches. If a vendor is selling you a no-touch agent that runs your business while you sleep, run. The good agentic systems are human-in-the-loop. They propose, you confirm. The pitch decks that hide that line are selling you risk.
2. The "let's connect 47 SaaS tools" dream. Every integration is a place where your agent will silently fail. Start with one workflow, one data source, one outcome. Earn the right to add the second.
3. Branded ChatGPT wrappers at $499/month. A lot of "AI tools for small business" are a system prompt and a logo. Before you buy, ask: what does this do that I can't get from a $20 ChatGPT seat plus 30 minutes of setup? If the seller can't answer in one sentence, pass.
The honest middle
The honest middle is unglamorous: a three-week build, a clearly defined workflow, a human checkpoint, and a measurable hour-saved-per-week number on the other side. That's the whole game right now.
Do that three times and you have a different business.