2026: The Year of AI Agents

February 15, 2026 · Updated October 2026

For a few years, AI at work mostly meant chat: you asked, it answered, and you did the work. This year that changed. The tools I use every day, Claude Code, GitHub Copilot and OpenAI Codex, can plan a task, use the systems it needs, and finish it, with a person checking the parts that matter. And it’s not just for engineers anymore. That’s why most of my work now is building AI agents for small businesses in New York.

Update, October 2026: I wrote this in February. Since then, the big companies went all in. xAI launched Grok Bot in August, Meta launched Muse in September, and OpenAI launched Dots a few weeks later. All three are agents with their own computer in the cloud that keep working after you close the app. The year of AI agents showed up early.

What an agent actually does

An agent is software that takes a goal, breaks it into steps, and works through them using your tools. The useful ones aren’t flashy. They’re narrow: one workflow, done the same way every time, with a log you can read afterwards.

The jobs I see most often:

• Qualifying new leads and booking the good ones
• Sorting the inbox and drafting replies to routine requests, which is exactly what an AI executive assistant does
• Pulling fields out of invoices, forms and contracts
• Answering common customer questions and handing off the rest
• Putting together the weekly numbers

Start with one workflow

The first agent teaches you more than any amount of planning. You find out which parts of your process are written down and which live in someone’s head, which data is clean, and which steps need a person. The second agent is quicker to build because of it.

Not sure which workflow to start with? That’s what AI consulting is for. We go through your week together and pick the one worth automating first.

I’d rather build one agent that gives a team back a few hours every week than a big system nobody trusts.

What solid custom AI agent development includes

• A clear job: one workflow with a known start and finish
• Your existing tools: it works in your email, CRM, calendar or database, not in a new app you have to log into
• Guardrails: a person approves anything risky, like paying an invoice or emailing a customer
• A record: every action is logged, so you can see what it did and why
• A bill you understand: the right model for each step, so you’re not paying top-model prices for simple sorting

None of this is science fiction. It’s careful engineering with tools that are good enough to use today.

“The future is already here — it’s just not evenly distributed.” — William Gibson

If there’s a workflow eating your week, tell me about it, or see how I build custom AI agents for New York businesses.