Everything just changed (and I'm going down the rabbit-hole)

Around a month ago, I got an email from Airtable.

If you’re not familiar with Airtable, they offer a table/database platform that’s like a cross between Google Sheets and a back-end database that developers might use, like Supabase.

I’ve been using Airtable for a while now for most of my clients, because it lets me set up fancy tables that all link to each other, fancy automations that use scripts to move data between tables and different platforms, and interfaces that let me give clients fancy dashboards and reports on their data.

If that sounds complicated already, think of it like this:

• All of your transactions in one place (whether they came from Teachable or Stripe or somewhere else)
• All of your leads and customers in one place (and linked to each of their transactions too)
• All of your products and pricing plans in one place (and linked to the people who’ve signed up for them)
• All of your automations in one place (like when a new transaction is added, a tag gets added in your email tool)
• And, since your data is now all in one place, super cool charts and reports that tie all of your data together too – instead of everything being spread across multiple platforms

On top of that, Airtable had rolled out a feature called “AI agent” fields, where instead of using a normal field type like text, number, or single select, you could insert a prompt for an AI model to process something in your data (translate a description into French, generate an image, search the web for context, write a snippet of code). The field would reprocess itself whenever the source data changed, so you could do loads of cool stuff with it.

Anyway, it turns out I’d been using this AI feature so extensively that I got labelled a “power user” by Airtable, which meant I got early access to an insane new tool they’d been working on behind the scenes – called Hyperagent.

First time inside Hyperagent

The email from the Airtable team said I’d been given early access to this platform with $1,000 in credits to play around with.

They were putting on webinars to show people the basics, so I signed up for one a few days in advance (no idea why, in hindsight; given that I usually don’t like webinars at all). But I joined the live webinar and just listened to it in the background while I signed up for an account and started seeing what it could do.

My first impressions were that it seemed cool, but I could see all the same things I didn’t like about other AI experiences. They were suggesting I start with one of their pre-made agents like the Chief of Staff or Data Analyst, but I read through what their roles were and knew they’d be a nuisance for me rather than particularly helpful. Sure, the concept of a Chief of Staff sounds good, but I don’t need more email summaries in my inbox and I had visions of corporate jargon and stand-ups and stuff that just wouldn’t go beyond yet another productivity performance.

Instead, I went to my personal inbox instead of my work one, hunting for actual jobs it could help me with – as opposed to just telling me what jobs I should be doing or what jobs I’ve already done (yawn).

Agent’s first steps

A few days old in my inbox, I found an email from Martin Lewis (Money Saving Expert) telling me in bright red lettering that I should urgently fix my tariff with my energy supplier if I haven’t already, because something about the war in Iran meant my energy prices would go up a lot higher soon if I didn’t.

I’d seen the email when it came in but I’ve been so busy recently that I couldn’t spare the head-space to figure out what he was advising, let alone go through all the admin of switching suppliers if that’s what he meant I should do.

This seemed a perfect test run of an agent, so I created one called Household Organiser, gave it access to my inbox, told it what supplier and tariff I’m on now, and said that I need help figuring out what to about this email (and with managing my household bills in general because I’m probably overpaying on some of them).

Huh, what was that?

On the surface, Household Organiser responded like a slightly more useful version of ChatGPT. It gave me a clear answer (yes, I should switch energy suppliers as soon as possible), explained why (something about the government’s price cap), and even listed out which suppliers were offering the best fixed rates right now, with links to each one.

But the output wasn’t the bit that made me go “huh” – it was what happened in the minute or so before I saw the output.

I could see this thing thinking about me.

There was this “logistical” running commentary at first, where I could see it:

• deciding to open up my Gmail since that was the only inbox connected;
• checking what date it was;
• searching for “Martin Lewis” within the last seven days;
• second-guessing itself(!) wondering if there might be multiple recent emails from him on the same topic;
• confirming via search that this was the only email that seemed relevant;
• reading the contents of the email;
• deciding to research the UK price cap and news about Iran to understand the context.

All very cool, but then it went further.

I saw it “thinking” things like:

Siobhán has said she’s been with [energy company] for around five years, and I’ve noticed they’re a green energy supplier so I should assume she wants to balance the ethical aspects of switching with any cost savings, even though she hasn’t mentioned that in this conversation. I’ve found cheaper rates with other suppliers, but let me check if her current supplier offers comparable rates and present them to her alongside the cheaper options.

Then:

Great, I’ve found a tariff with Siobhán’s current provider that is only 4.2% more expensive than the cheapest from other companies. This tariff is also 5.4% cheaper than her current variable tariff and it won’t involve the kind of admin that Siobhán implied she doesn’t have bandwidth for at the moment. Let me research the exact steps Siobhán would need to take through her current online portal, then present my findings to her. I’ll present all of the options but make clear the assumptions I’ve made about her reasons for being with a green energy supplier.

Like, what?

It’s learning about me, fast

This “thoughtfulness” blew my mind from the start, but it kept compounding as it walked me through the various steps in this test task.

Once I’d confirmed it made the right call about fixing rates with my current supplier, it seemed to learn that as a “thing” about me – i.e. that I prioritise reducing admin over saving every penny. From then on, I could see it thinking about how to minimise the steps it would take for me to do something, even if it meant paying a small amount for a tool or someone to do something for me.

Within 15 minutes, I’d switched my tariff, booked an engineer to install a smart meter… and then the agent went a step further by asking if I wanted it to keep track of when my tariff was going to end and look for other options at that point (in a couple of years!).

I asked myself “how could I get it to actually do that?” (ever-aware that unless I set up some systems or schedules for this thing, it can’t actually do anything unless I nudge it into life each time). Then I started to get something.

Can you do this?

I grilled Household Organiser for the next 10 minutes.

Can you connect to Airtable’s API?
Yes.

Can you create tables and fields and update them?
Yes.

And you can run on a schedule with specific prompts?
Yes.

And I could even just email you to get you to do something?
Yes.

So if I forwarded my household bills when they come in, you could start keeping track of stuff in an Airtable table that you’re responsible for building and maintaining?
Yes.

And you could check that table on a regular basis to decide if there’s anything you should research or flag to me?
Yes.

Wait, and another agent could access that same table if I made like a “Personal Finance Advisor”?
Yes.

Okay, I just gave you access to your own Airtable base. Can you build what you need now?
Yes.

And I watched this thing build its own tables, lists, workflows, and schedules – without me touching any back-end UI or writing a single line of code.

ChatGPT, eat your heart out.

I just got replaced

Even from that first session (which I now know doesn’t scratch the surface of what’s possible with these things), I could see this was huge. Not because it helped me with a task as an end user, but because it replaced me as a professional tech lead.

This kind of technical automation and back-end system work is what I’ve done for clients for years. They tell me they want to do XYZ and I build the tech that lets them do it.

But Household Organiser built a back-end system and automation flow for managing household bills in 10 minutes, where it might have taken me hours to hook up for a client. Not only that, but I could see it designed the architecture well too.

So I saw two choices:

1. Bury my head in the sand, hope my clients don’t find out about this any time soon, and use this to benefit my own work and personal life, kind of in secret, before finding a new career…
2. Or see how far this thing goes and be the one jumping in head first.

If you know me even slightly, you’ll know it wasn’t even a choice. If curiosity killed the cat, you might as well call me Whiskers. I’m a month down the rabbit-hole and I haven’t looked back since.

Learning (and unlearning) in public

You’re reading this as a blog post because for the first time in my life, I have something I want to blog about. I’ve had blogs before to try and get traffic for various projects, but that’s just “content”, right?

With this, I can’t even begin to explain how fast my understanding of AI agents is evolving every day because of the speed at which these systems can be developed and deployed.

I’m learning whole new ways of thinking. It turns out that applying old-world approaches to building technical infrastructure actually just makes these self-learning agents brittle and ineffective – and surprisingly the architecture matters more here, not less.

I’ve also realised there are orders of magnitude more ways agents can open up massive vulnerabilities for online businesses if they’re not given guardrails, so I’m diving into that for my clients as we speak.

All that to say:
Everything just changed; whether you realise how it will affect you yet or not. I don’t think it’s a stretch to say that this is a second industrial revolution.

And if you want to come along for the ride, I’ll be waffling about everything I’m building and learning right here. Feel free to jump into the Agent Apocalypse with me.

More blog posts