What if AGI is already here? Will Humphrey cut UK bureaucracy overnight? And what should (and shouldn't) you be doing with AI right now?

Written by Fola Yahaya

What if AGI is already here?

I’m not a conspiracy theorist, but in these troubling geopolitical times, I find myself increasingly open to previously improbable theories. On the AI front, it's worth considering the possibility that the AI we're currently experimenting with is vastly inferior to the mega cluster-powered AI systems available only to the world's most powerful corporations and governments. Take Google, for example. It has access to our collective digital footprints – our searches, locations and preferences – and virtually every website is crawled by its bots. This illustrates the enormous power that Big Tech companies can wield when combining vast datasets with millions of processing chips.

In the video above, political commentator and journalist Ezra Klein and Ben Buchanan, a former AI adviser in the Biden White House, discuss the growing belief that superintelligent AI – systems potentially thousands of times smarter than ChatGPT – are either already here or rapidly approaching. Especially in the Trump/Musk era, AI is shifting from hype to reality, quickly becoming central to governance, society and defence. From Musk’s controversial proposals to replace civil servants with AI, to Big Tech's pursuit of increased defence spending, we're rapidly approaching a world dominated by AI.

So what does this mean for us? The uncomfortable truth is that nobody really knows. Just as the inner workings of today's advanced AI remain somewhat opaque, the societal impacts of achieving true AGI are impossible to predict with certainty – particularly regarding labour markets, national security, cyber operations and societal structures. The leading AI research firm Anthropic recently identified "early warning signs" of AI systems exhibiting worrying capabilities in cybersecurity and biology, highlighting vulnerabilities in current alignment strategies. Miles Brundage, former policy lead at OpenAI, anticipates a "very serious incident" involving AI within the year, potentially causing billions in damage or even loss of life.

We do indeed live in interesting times…

The problem at the heart of AI in governments

Taking a leaf out of Elon Musk’s DOGE playbook, the UK Government just announced plans to ‘mainline’ AI into every governmental process and drive annual savings of £45 billion. At the heart of this AI roll-out is Humphrey, an AI system that it is hoped will cut red tape and enable UK civil servants to be more efficient at executing processes (and ultimately automate themselves out of a job).

Whenever I hear of attempts to slim down bureaucracy I’m reminded of my stint as a very expensive management consultant. Even though I was but a freshly minted PhD with very little, if any, knowledge of the ways of the working world, I was regularly charged out at a ridiculous £1,200 per day. One of my first gigs was to work out how to re-engineer the antiquated UK Government's ministerial correspondence process. Back in the 2000s, if anyone wrote a letter to a minister, the response had to be drafted by ministry officials, printed out for physical review and signature, then posted and routed through various departments for multiple rounds of sign-off. Alongside a team of five equally expensive consultants, I spent three months holed up in a windowless office literally mapping this process and in the process learning about how massively inefficient and costly it was. Needless to say, we were quick to recommend a digital solution to this stupidly manual process. I was off to my next consulting gig before the lengthy and equally inefficient procurement process to select a vendor was concluded, but I can guarantee that the final solution was just a digital version of a paper-based process.

Herein lies the problem. Rolling out AI to fix broken systems misses the point entirely. Instead of fixing or streamlining broken processes, AI becomes a shiny plaster over a gaping wound. The result? The same old problems – just processed quicker and spread wider. It’s also ironic that the UK Government has named its core AI system after a fictional civil servant from a long-running 1970s comedy series called Yes Minister. In this sitcom, Sir Humphrey is a senior civil servant notorious for preventing meaningful change, fiercely protecting bureaucratic traditions at all costs. Naming the heart of their digital transformation after such a figure suggests either a staggering lack of self-awareness or perhaps a knowing nod to the inertia deeply rooted within governmental bureaucracy.

You'd be hard-pressed to find any taxpayer (unless they're a civil servant) who thinks their government shouldn't be slimmed down and automated. But just as Elon Musk is discovering with his DOGE-fuelled ambitions, fighting the Sir Humphreys who resist the all too necessary cuts to civil service bloat will take more than AI – it requires fundamentally reimagining broken systems rather than simply digitising inefficiency. Without tackling the underlying bureaucratic culture that Sir Humphrey so perfectly embodied, the UK Government's £45 billion savings target risks becoming yet another expensive consultation exercise that produces impressive PowerPoints but minimal change. The irony of naming their AI system "Humphrey" might be the most honest admission yet about the true challenges ahead.

What you should and shouldn't be doing with AI right now

As the 'AI guy' in my circle, I'm frequently asked about practical AI use rather than just theory. I’m painfully aware of how easy it is to get carried away and waste time and money on the latest new AI app or gadget. With subscriptions and token use costing around $50 per month, owning multiple tools quickly becomes expensive. Even worse are the seven day free trials that you forget to cancel and that continue their digital shakedown in the background.

So I thought it would be useful to track exactly how AI tools have either helped or hindered me this week. Below is a snapshot of my AI experiments:

Building a mobile app to take a picture and assess the amount of net carbs to allow me to correctly calculate an insulin dose for my son

Process: Signed up to Replit (a mobile, no-code app). Asked the AI bot all of six questions, plugged in my Google AI Studio API key to allow the AI to use my camera to assess carbs directly from images.

• Time: 10 mins
• Cost: Free
• Worth it? ✅ Absolutely – we now use this every day!

Converting Adobe XD animation to code (v0.dev)

• Process: My designers created an animation of Strategic Agenda’s services in Adobe XD, but it didn’t play well in our emails to clients. I uploaded a screenshot into v0.dev; 20 chat messages later, I'd recreated in 20 minutes what originally took a designer three days.
• Time: 20 mins
• Cost: Part of a $20 subscription
• Worth it? ❌ No, the export quality was poor, but an interesting test of the tool's current capabilities.

Uploading my son’s exam timetable to Gmail (OpenAI Operator)

• Process: Connected OpenAI Operator to my Gmail and uploaded a PDF timetable. Unfortunately, the AI incorrectly populated calendar entries, which I had to fix manually.
• Time: 2 hours
• Cost: Free
• Worth it? ❌ Definitely not. A colossal waste of time.

Building a prototype of a client data visualisation dashboard (v0.dev)

• Process: Used v0.dev extensively for a quick data visualisation dashboard prototype.
• Time: 4 hours
• Cost: Part of a $20 subscription
• Worth it? ✅ Absolutely – a highly effective prototype.

Creating a Gmail add-on for CRM integration (Claude and Google Scripts)

• Process: Built a Gmail add-on that instantly sends leads directly into our CRM. Previously, this took three weeks, cost $1,500 and frequently malfunctioned. This AI-driven approach was faster, free and reliable.
• Time: 2 hours
• Cost: Part of a $20 subscription
• Worth it? ✅ Definitely – saved money and it delivered perfectly.

Automatic meeting notes (Bubbles and Gemini AI)

• Process: We previously used an app called Bubbles (free version) for automatic transcription, summaries and follow-up actions for meetings. However, we’ve now switched to Google’s inbuilt minute taker which provides the same functionality at no cost when invited to a Google Meet.
• Time: Automatic (minimal set-up)
• Cost: Free
• Worth it? ✅ Definitely – a no-brainer upgrade.

Creating a video for this newsletter (Kling AI, ElevenLabs, Midjourney)

• Process: Created a short video to accompany my article on AI in governments at the start of this newsletter using Kling AI for video creation, ElevenLabs for voiceover and Midjourney for images.
• Time: 1 hour
• Cost: Free
• Worth it? ⚠️ Probably not (7/10 result), with the caveat that it’s good to see how far AI video production has come and has left to go before it’s fully production ready.

Creating a Studio Ghibli-style image of myself

Process: Uploaded a reference photo and asked ChatGPT to generate a Studio Ghibli-style portrait.

Time: 30 secs

Cost: Free

Worth It? ⚠️ Not really, but fun nonetheless.

Weekly totals: Time spent: 9.5 hours | Cost: $40 plus partial subscription fees

Hopefully these examples inspire you to try AI and understand its potential as well as its limitations.

AI robot video

Boston Dynamics released footage of its latest updated Atlas humanoid robot. Using reinforcement learning and with references from human motion capture and video, Atlas isn't far off being battle-ready...

What we’re reading this week

• OpenAI launched its “most advanced image generator yet”.
• Starting 1 September, elementary and secondary students in China will get at least eight hours of AI training per academic year, either as a standalone subject or integrated into science/IT curricula.
• BYD has unveiled a new charging system that can add about 470 km of range in just five minutes – potentially leapfrogging Tesla’s supercharger network.
• DOGE’s plans to replace humans with AI are already under way.

That's all for this week. Subscribe for the latest innovations and developments with AI.

So you don't miss a thing, follow our Instagram and X pages for more creative content and insights into our work and what we do.