Where Will AI Be in 2 Years? What Business Owners Need to Know

Right now, AI agents are the hot thing. Everyone is talking about them. By 2028, nobody will talk about them because they will just be how things work — like nobody talks about “using email” anymore.

Here is a concrete example. One of our clients receives about 40 enquiries a day through their website. Today, an agent reads the enquiry, researches the prospect on LinkedIn, qualifies them against a scoring rubric, drafts a personalised reply, schedules a follow-up, and logs everything in HubSpot. A human reviews the top-tier leads. The rest are handled end-to-end.

That system took weeks to build and required careful prompt engineering. By 2028, you will spin up something equivalent in an afternoon. The agent frameworks will be mature. The models will be better at multi-step reasoning. The cost per run will be negligible. And the businesses that already have their data pipelines and integrations in place will bolt on agent capabilities like adding an app to a phone. Those starting from zero will still be figuring out how to connect their CRM to anything.

Multimodal means the AI handles text, images, audio, and video together. Today it works, but it is expensive and a bit clunky for production use. By 2028, it will be cheap and reliable enough to throw at problems you would never consider automating right now.

I work with a building company that currently pays someone to review site photos and write up defect reports. Tedious, slow, expensive. Within two years, a model will look at those photos, identify the issues, cross-reference the building code, and produce a compliant report — for cents. That is not science fiction. The models can mostly do it today. They just are not reliable or cheap enough yet for production. They will be.

The same logic applies to phone calls, voicemails, handwritten forms, video walkthroughs, and anything else that currently requires a human to watch, listen, or read. Real-time processing will make this even more powerful — AI that responds in milliseconds changes what is possible for customer-facing tools. Voice AI that handles inbound calls naturally. Live chat that is genuinely indistinguishable from a human. These already exist in early forms and they will mature fast.

API costs have dropped roughly 10–50x in two years. That trend is not slowing down. Open-source models are closing the gap with proprietary ones at speed. By 2028, running a capable model locally will cost essentially nothing.

This changes the maths on every automation decision you have ever made. We regularly tell clients “that process is not worth automating yet — the volume is too low for the AI spend to make sense.” Within two years, nearly every one of those edge cases will flip to being worth it. A task that costs $0.10 per run today will cost $0.005. Suddenly, automating a process that runs 20 times a day saves real money.

Here is the part people miss. If you already have your automation infrastructure built — the workflows, the integrations, the data pipelines — you expand almost for free when costs drop. You swap in a cheaper model. Done. But if you have nothing built, you are starting from scratch while your competitors are just adjusting a setting.

This is the bit that matters most and gets the least attention. AI automation is not a linear improvement — it compounds.

Every workflow you automate generates data. That data makes the next automation smarter. Every integration you build becomes a building block for the next one. Your team develops AI literacy by working alongside these systems daily. After twelve months of steady building, you are not 12x ahead of where you started. You are 50x ahead, because each piece accelerated the next.

A competitor who starts two years late does not just need to build what you built. They need to build it without the data, the institutional knowledge, or the team skills you developed along the way. That gap is extremely difficult to close with money alone.

I hear this constantly: “We will wait for the tech to mature.” That logic made sense with previous technology waves. You could wait for smartphones to get good, then adopt. You could wait for cloud computing to stabilise, then migrate. The tech did not compound while you waited.

AI is different. Every month you wait is a month of wasted labour spend you will never get back. It is a month your competitors are refining their systems while yours do not exist yet. It is a month of falling further behind a moving target.

There is a talent squeeze coming too. Good AI implementation partners are already scarce. As demand explodes over the next two years, the best ones will be fully booked. The businesses that lock in relationships now get priority. Everyone else queues.

Enough theory. Here is what I would do if I were running a small or mid-sized business right now and wanted to be in a strong position by 2028.

Pick your most expensive manual process and automate it properly. Not a proof of concept. Not a pilot. A real production system that replaces real labour hours. Use open tools like n8n and Supabase so you own everything. Get it live within 30 days and measure the ROI. That first win funds and justifies everything that comes after.

Design every system to be model-agnostic. The model you use today will be obsolete within months. Build your workflows so the AI layer swaps out cleanly. When something better and cheaper drops, you change a config value — not rewrite the whole system. That is the core of a good AI integration strategy.

Get your team comfortable with AI now. They do not need to become engineers. They need to understand what AI can do well, where it fails, and how to spot processes ripe for automation. The best automation ideas I have seen come from frontline staff, not executives.

Find an AI partner who stays involved. This space moves too fast to track on your own. You need someone who watches every model release, tests what actually works, and brings you upgrades proactively. Quarterly reviews, prompt refinements, new capability assessments. We wrote about this in our guide to future-proofing your business with AI.