April’s been a bit different. No podcast episode this time. A combination of holidays and illness has kept Reece and I out of the studio this month, and I refuse to record an episode that sounds like it was made in a garden centre car park in a hurry. But the AI world didn’t slow down to wait for us, and I’ve got more to cram into this one edition than most.

Think of this as April’s full debrief.

Before the news: something I’m genuinely excited to share.

🔥 The Big Story

We’ve Just Launched Our First Open Learning Course

After running our Open Learning survey earlier this year, where I asked Somerset businesses what skills they actually wanted to develop, one answer came back loud and clear: automation. Not AI in the abstract. Not prompt engineering. Automation. The stuff that quietly eats hours out of every working week, and that people know they should be doing something about but haven’t quite known where to start.

So we’re starting there.

Automation Fundamentals is our first Open Learning course, and it is now live and taking bookings. It is a six-hour, in-person day at iAero in Yeovil, designed to give you a proper foundation in how automation works, without locking you into any single platform.

The first cohort runs on 1 June 2026, with a second on 1 July 2026. First cohort bookings are at the introductory launch price of £150 + VAT (the standard price will be £200 + VAT).

Here’s what makes this different from a software tutorial:

It is tool-agnostic by design. We cover Power Automate, Make, Zapier, n8n, relay.app, and IFTTT, but the focus is not on clicking through menus. It is on teaching you the language underneath all of them: triggers, actions, conditions, loops, approvals, and error handling. The bits that don’t change no matter which tool you end up using.

You’ll also get early access to AutonoSim, a virtual automation builder I’ve built specifically for this course. It lets you design and test automations safely, with AI built in to spot problems, before you go anywhere near a live system. No paid subscriptions. No setup friction. Just practice.

At the end of the day you will leave with a Techosaurus digital certificate, a LinkedIn badge, and a set of processes you can actually implement.

This course isn’t just for the people who will build the automations. It is equally for the decision makers, the process owners, and the leaders who keep being told “we should automate this” but aren’t quite sure what that means or how to specify it. If you understand how automation works, you can have better conversations, make better decisions, and stop being at the mercy of people who tell you it’s all very complicated.

Automation is becoming basic business literacy. The organisations pulling ahead are not necessarily the most technical. They’re the ones that understand process, flow, and where humans should stay in the loop.

Book 1st June 2026 — £150 + VAT

Book 1st July 2026 — £200 + VAT

See all course dates and details

📰 Other News

ChatGPT Just Had Its Biggest Month in a While

April was remarkable for OpenAI. In the space of three days they shipped three significant things.

GPT Image 2 / ChatGPT Images 2.0 landed on 21 April. The headline is near-perfect text rendering — accuracy above 95% across Latin, Chinese, Japanese, Korean, and Arabic scripts — at 4K resolution. It also has thinking capabilities, meaning it can search the web, generate up to eight coherent images from a single prompt, and self-check its creations. If you’ve been frustrated by image generation that cannot spell, or that loses consistency across a batch of images, this is a genuine step forward. (Source: OpenAI, 21 April 2026)

Workspace agents arrived on 22 April. Codex-powered team agents that take on complex, repeatable workflows inside ChatGPT Business, connecting to Slack, Salesforce, email, calendars, and more. OpenAI is positioning these as the direct successor to Custom GPTs, and they are right to do so. I have written a longer piece on workspace agents and there’s more from me in the Soapbox section below. (Source: OpenAI, 22 April 2026)

GPT 5.5 followed on 23 April — OpenAI’s most capable model to date, with particular strength in agentic tasks, coding, and multi-step work. I’ve been running it alongside Claude Opus 4.7 for research and writing tasks this week, and the competition is genuinely tight again. Which is exactly how it should be. (Source: OpenAI, 23 April 2026)

ChatGPT also launched for CarPlay this month and added a dedicated Spreadsheets app. Both are worth looking at if you use either in daily life — the spreadsheets integration in particular is a significant addition if you live in Excel or Google Sheets.

Claude Opus 4.7 and a 24-Hour U-Turn

Anthropic released Claude Opus 4.7 on 16 April. It is a meaningful upgrade, particularly for advanced software engineering and complex tasks. It is also the first Claude model with high-resolution image support, capable of processing images up to 3.75 megapixels, up from 1.15MP in Opus 4.6. Pricing stays the same. (Source: Anthropic, 16 April 2026)

But April was not entirely smooth for Anthropic. On 21 April, some users noticed that Claude Code had disappeared from the Pro plan pricing page. No announcement, no email, no changelog entry. Just a red X where it used to be. Reddit, Hacker News, and X lit up with complaints within hours. By the morning of 22 April, it was back.

Anthropic described it as “a small test of 2% of new signups,” which did not land particularly well as a framing for something that read as a silent downgrade. Alongside this, Anthropic also acknowledged several Claude Code performance issues from recent weeks: reduced reasoning effort that had been quietly dialled down in March, a bug that was discarding reasoning history mid-session, and a system prompt change capping response length to 25 words between tool calls.

They have since fixed all three. The lesson, as ever: the community notices, and the community will tell you. (Sources: The Register, Fortune)

Google Workspace Intelligence

Google launched Workspace Intelligence on 22 April, announced at Google Cloud Next, and it is a bigger deal than the name suggests.

Up until now, Gemini inside Workspace has been helpful but essentially stateless. It would assist you with a document, but it didn’t really know what was in your business. Workspace Intelligence changes that. It builds a real-time semantic layer across Gmail, Drive, Docs, Sheets, Chat, and Calendar — and makes that context available to Gemini automatically, without you having to paste anything in.

Ask it to draft a project update and it will already know the relevant emails, meeting notes, and collaborators. Ask it who you have been talking to about a specific topic and it will surface that from your inbox. It is grounding Gemini in your actual business knowledge rather than requiring you to bring that knowledge to it.

Worth enabling if you are on a Google Workspace plan. The data stays in your tenant, and Google has committed to not using it for ad targeting or model training. (Source: Google Workspace Blog, 22 April 2026)

M365 Connector Now Free for All Claude Users

A quiet but significant expansion this month: Anthropic’s Microsoft 365 connector for Claude is now available on all plans, including the free tier.

This means Claude can now read your Outlook emails, search SharePoint and OneDrive documents, and pull from Teams conversations — on any plan. Previously this was restricted to higher-tier subscriptions. If you are in a Microsoft 365 environment and you have been using Claude separately, this is the bridge that connects the two. (Source: UC Today)

Quick-Fire: A Few More Things Worth Knowing

Copilot Critique: Microsoft has quietly begun using Claude alongside GPT-4 inside Copilot for certain tasks — pairing Anthropic’s model as a checking layer on outputs. Interesting architecture, and a sign that even inside the big ecosystem plays, models are being mixed and matched on merit rather than vendor loyalty.

Gemini Mac App: Google launched a native Mac app for Gemini this month. If the browser version has felt clunky, it is worth downloading.

Perplexity Personal Computer: Perplexity has rolled out an early version of a personal computer agent — software that can operate across apps on your desktop. Another signal that “AI that runs on your machine” is moving from experiment to expected.

OpenAI Codex expands: Codex’s ability to control computer tasks has been extended beyond coding. The line between AI assistant and autonomous operator continues to blur.

Claude Design: Anthropic has launched a new Claude Design product aimed at creative and design workflows. Still early days, but worth watching.

💬 Scott’s Soapbox

Skills and Workspace Agents Have Killed the Custom GPT

If you set up a Custom GPT a year ago and haven’t really used it since, I want you to know: you were not wrong about the idea. You were just ahead of the implementation.

Custom GPTs were a promise. Skills and workspace agents are the delivery.

Skills in ChatGPT are modular units of expertise. You build one skill for one thing — writing in your voice, generating images in a specific style, formatting a particular kind of document — and the AI loads it automatically when it is relevant, or you call it explicitly. The critical upgrade is the stacking. With Custom GPTs you had to pick one. With skills you can combine them. My content workflow now chains a voice skill, an image style skill, and a social media post skill into a single agent. I say “do it all.” I come back five minutes later.

Workspace agents extend this further. They are scheduled, they have access to real tools — email, calendar, Slack — and they run autonomously in the cloud. My morning briefing agent reads my inbox and calendar every weekday at 6.30am and sends me a prioritised brief. I gave it permission to send one email. That is the only thing it is allowed to do. Everything else is read-only. The guardrails are explicit and deliberate.

That is not a chatbot persona. That is operational infrastructure.

OpenAI has positioned workspace agents as the direct successor to Custom GPTs. That framing is accurate. If you are still thinking in Custom GPT terms, the mental model needs updating.

I’ve written a longer piece on how skills and agents work in practice, and why the right starting question is not “what’s the difference between all these things” but “what repeatable job in my week could an agent handle?”

💡 Try This

This week’s challenge: build your voice skill in ChatGPT.

Here is the process that works:

1. Interview yourself. Open a chat in ChatGPT and ask it to interview you about how you write — what topics you cover, what phrases come naturally, what you would never say, and what you want readers to feel when they read you. Do not rush this. Let the conversation breathe.

2. Ask for a warts-and-all document. At the end of the interview, ask it to produce a comprehensive voice profile — how you write, what makes it distinct, what to avoid.

3. Test it against something off-topic. Go to a news site. Find a story you have no professional interest in — something completely outside your normal territory. Give it to the AI and ask it to rewrite the story in your voice using the profile. Read it back and note anything that jars. That is where the skill needs refining.

4. Build the skill. In ChatGPT Business, go to /skills and create a new skill using “create with chat.” Give it the voice profile document as a reference file. Let it build the skill and test it before installing.

5. Refine over time. Every time the voice is not quite right, go back and update the skill — tell it what missed, and it will update the underlying files. The skill gets sharper the more you use it.

Once it works well in ChatGPT, you can download the skill as a zip file and install it in Claude too.

📆 Digital Hub

The next Digital Hub in Yeovil is coming up shortly, and I am looking forward to this one.

Shane Evans — qualified hypnotist, life coach, and one of the more unusually useful people I know — is talking about something we are not discussing enough in the AI conversation.

AI amplifies everything. Including the bad stuff. People are starting to burn out as their output capacity doubles or triples overnight and the goalposts move to match. Shane’s session is going to look at how we keep AI as the tool and not let it run us.

That is a conversation worth having, and if you can be in Yeovil for it, it is worth being in the room.

Keep an eye on yeovildigitalhub.co.uk for dates and booking.

Scott Quilter | Co-Founder & Chief AI & Innovation Officer, Techosaurus LTD