AI agents you can trust, and why big AI tools change the game for small firms
How small businesses can use AI agents practically, why AI-written code is now normal, and what specialized AI tools mean for you.
Summary
- AI agents work better when you stay the manager — Ethan Mollick shows a supervised agent beats one you “hand everything to”.
- AI-written code is entering even the strictest places — Debian won’t ban AI code, meaning the tools your business runs on are already built with AI.
- AI is becoming a specialized industry tool — new assistants are being built for specific professions, showing where small-business tools are heading too.
Core feeling: AI is no longer a universal magic tool — it’s becoming a daily instrument that needs your supervision and the right application.
1. AI agents work better when you stay in charge
What changed. Ethan Mollick’s new piece Agency and Agents explains something important: AI agents — tools that carry out several steps on their own — produce better results when a human stays involved and directs the process, rather than handing over the whole task and walking away.
What it means for a small business. If you’re the only person in your business and hoped AI would “do it all itself”, this is an important correction. An agent isn’t an employee you hand a task and forget. It’s like an intern — capable, fast, but needs watching. If you run a small accounting office and ask AI to “organize all client documents”, you’ll get a mess. But if you ask step by step — “sort these invoices by month”, check, then “now total the amounts” — the result is reliable.
What you can do this week.
- Take one real task (e.g. the week’s invoices or 5 customer email replies) and try it with ChatGPT (free version or €20/mo) or Claude (free version or €18/mo) — split it into steps, check each one.
- Don’t ask for one giant task — ask for 3-4 small steps in sequence and correct as you go.
- Note where AI got it wrong — that’s how you learn which jobs it’s good for and which not.
What to watch.
- Over the coming months agent tools will get simpler, but the need for supervision won’t disappear — don’t trust promises of “full automation without a human”.
- Tools that let you approve each step (rather than running everything at once) will be the safer choice for a small business.
2. AI-written code is already in your tools — even in the strictest places
What changed. Debian, one of the oldest and most conservative Linux systems, decided not to ban AI-written code in its software. That means even the strictest software communities now accept AI as part of writing code.
What it means for a small business. You probably never thought about how your tools are written — accounting software, your online store platform, a booking system. But this news means one practical thing: AI-written code is now completely normal, and that opens up to you too. If you run a 5-person workshop and always thought “I’d love a small tool that counts materials, but a developer is too expensive” — now you can try to build it yourself with AI.
What you can do this week.
- Try Replit (free mode) or Lovable (free to start) — describe a simple tool in plain language and it generates a working version with no coding knowledge.
- Start with a small problem: a calculator, a simple customer form, a list tool — not your whole business system at once.
- If it works, show it to someone who understands coding to check security before using it with real customer data.
What to watch.
- Building tools “in your own words” will get more accessible — within a year small businesses will have built tools that today require an agency.
- Be careful with customer personal data in AI-built tools — verify data security before using them for real.
3. AI is becoming a specialized industry tool — not a universal chat
What changed. More and more AI tools are being built for a specific profession rather than “everything”. For example, a new assistant for police officers raised $6 million in funding — modeled on Harvey, built for lawyers. The trend is clear: AI is moving from universal chat to specialized work tool.
What it means for a small business. Universal ChatGPT is a good start, but the future is tools that know YOUR industry exactly. If you’re a hairdresser, accountant or craftsman, soon there will be tools that understand your industry’s terms, rules and typical situations without long explanations. Practically, you can already build yourself a “specialized” assistant by training a universal tool with your own information.
What you can do this week.
- ChatGPT (€20/mo) lets you create a custom “GPT” — feed in your most common questions, price list, processes, and it answers in your industry’s context.
- Claude Projects (€18/mo) — upload your documents in one place and the assistant answers based only on your materials.
- Start with one frequently repeated task — preparing quotes, customer replies, price calculations.
What to watch.
- Over the coming year, cheap industry tools will appear for smaller professions — watch for when someone builds one for yours.
- Universal tools (ChatGPT, Claude) will stay useful, but specialized ones will save more time on specific jobs.
Today’s picture
The unifying change: AI is ceasing to be a “magic box you toss a task at” and becoming a daily instrument that needs your direction and the right application. Mollick says — stay the manager. Debian shows — AI is already in your tools, including ones you can build yourself. Specialized assistants show — the future is specific jobs, not general conversation. In all three cases, AI is more valuable the more precisely you use it.
| Event | Consequence |
|---|---|
| Mollick: agents work better with human supervision | Split tasks into steps, check each — don’t trust “full automation” |
| Debian won’t ban AI code | AI-built tools are the norm — you can build a simple tool yourself without a developer |
| Specialized AI assistants for industries | Build yourself a custom assistant with your own information now |
3 questions for a small business owner:
- Which repeated task in your week takes the most time — and have you tried breaking it into steps with AI?
- Is there a small tool you always wanted but a developer was too expensive — could you try building it yourself?
- Where are you already letting AI run without checking, and is that safe with your customer data?
Sources
AI news for business leaders
Twice a week: what AI news means for a small business. No hype, with concrete steps.