The Algebra of AI Organisations

Friends,

Welcome to The AI Optimist - your weekly dose of AI optimism. Here's what matters…

2026 has arrived. CES is underway in Las Vegas, everyone's talking about Claude Code, and a question keeps surfacing in our conversations: what shape should your organisation take when AI changes the cost of everything?

But there's another question we're hearing even more often, especially from parents: "What will my kids do for work?" It's the anxiety lurking beneath every AI conversation - if machines can do cognitive work, what future are we preparing the next generation for?

Here's our take: the teenagers who thrive won't be the ones who avoided AI - they'll be the ones who learned to work alongside it early. That's why we're excited that Startup Sherpas has just launched paid work experience for 13-19 year olds where they learn AI skills, get real work experience with known companies, and get paid cash for doing so.* Parents can sign their kids up directly. If you're worried about your teenager's future, this is one concrete thing you can do about it.

Now, onto this week's big idea.

Let's get into it.

🔥 URGENT PRIORITIES

✅ No immediate fires - CES announcements are forward-looking, not requiring action today

✅ Claude Code and similar tools are available now - worth experimenting, not panic-adopting

✅ Open models continue evolving rapidly - strategic positioning, not crisis

Use the calm to think structurally about your organisation design.

🎯 STRATEGIC CONSIDERATIONS

Organisational Algebra: The Foundation for an AI Operating System That Doesn't Hallucinate

Ben has been developing what he calls "Organisational Algebra" - a framework for making AI genuinely useful in business operations. He's shared his thinking in two recent LinkedIn posts (the concept and a practical example).

The core insight is counterintuitive: most AI implementations fail because they give models too much freedom. When you ask an AI to "analyse this data" or "improve this process," you're essentially saying "here's some context, figure it out." The AI might produce something impressive-sounding, but it's often hallucinated nonsense that doesn't connect to how your business actually works.

What if instead, you gave AI a structured language that mirrors your business logic?

Here's the practical example Ben demonstrated. He described a simple business - a coffee shop - in plain English. But instead of asking AI to "create a business plan," he gave it a structured framework where:

• Every business process has defined inputs (coffee beans, staff time, equipment) and outputs (drinks sold)

• Outputs have prices; inputs have costs

• Processes connect together (making coffee → serving customers → taking payment)

• The numbers roll up automatically

The result? From that single description, the AI generated two things simultaneously:

1. A process diagram showing how a customer visit flows through the shop

2. A financial forecast - not just revenue, but proper P&L, cash flow, and balance sheet

Same description. Same underlying logic. Two different views that stay perfectly in sync.

Why this matters for your business: The AI didn't need to understand accounting or guess at your margins. The structure handled that. It just had to describe valid business operations - and the framework ensured everything connected properly.

This is what Ben means by "algebraically constrained" AI. Give models concrete rules instead of open-ended prompts, and they stop making things up. They start building things that actually work.

The bigger picture: This isn't just a clever technique - it's the foundation for an AI Operating System for your business. One that leverages the power of both code and AI: the creativity and flexibility of language models, combined with the precision and reliability of software. You get the advantages of AI with none of the hallucination risks. The structure acts as guardrails, ensuring AI can only produce outputs that are logically valid and financially sound.

Your move: Before your next AI initiative, ask: "Are we giving AI structure or vibes?" If you're hoping the model will figure out your messy processes, you're setting it up to fail. If you can define clear inputs, outputs, and how they connect, AI becomes remarkably useful. If you want to explore this approach for your actual business, Ben's DMs are open.

🤓 GEEK OUT

🎧 The Audio Interface Is Coming

👉 TechCrunch

OpenAI is betting big on audio AI, unifying engineering teams for an audio-first personal device expected in about a year. Jony Ive, who joined through OpenAI's $6.5 billion acquisition, sees audio-first design as a chance to "right the wrongs" of past consumer gadgets.

Why this matters: Your audio strategy matters. As AI moves from screens to voice, the organisations best positioned will be those who've thought about voice-based workflows and ambient computing. Start experimenting with voice interfaces now.

🏗️ Context Engineering: Lessons from Building Manus

👉 Manus blog

Manus (recently acquired by Meta) published a deep-dive on "context engineering" - designing what information AI agents receive. Key insight: keeping failed attempts in context actually improves performance. The agent learns from its mistakes within the session.

Why this matters: "Prompt engineering" is evolving into "context engineering" - a more sophisticated discipline about designing the entire information environment for AI systems. Early practitioners of this skill will have significant advantages in building effective AI workflows.

📈 8 Plots Showing the State of Open Models

👉 Interconnects

Nathan Lambert's data shows Chinese open models now dominate every adoption metric. Qwen's December downloads exceeded literally every other organisation combined. The only opportunity for Western models may be at the large-scale frontier.

Why this matters: If you're building on open models, understanding this landscape is essential. The default choice for many applications is now Chinese-developed, which has implications for supply chain, compliance, and long-term strategy.

🎨 WEEKEND PLAYGROUND

Try "Claude Code" (A Terrible Name for a Transformative Tool)

👉 Read Transformer's take

Let's be clear: "Claude Code" is a terrible name. It makes you think this is for developers. It isn't.

Code is just a language by which we instruct computers to do things. An AI that can write and execute code can do almost anything you do on a computer. The question isn't "is this a coding task?" It's "can this be done digitally?"

Shakeel Hashim at Transformer has zero coding experience. In two weeks, he's had Claude Code prepare his tax filing from bank statements (it got everything right), book theatre tickets by checking his calendar against venue availability, and build automation tools saving his team half a day per week.

Try it yourself:

✅ If you have any terminal familiarity: install Claude Code and start with "Organise my downloads folder by file type"

✅ If you don't: ask a technical colleague for a 10-minute demo - watching it will reshape your understanding of what's possible

✅ Graduate to something useful: "Go through these invoices and create a summary spreadsheet"

The barrier is knowing basic terminal commands - not hard, but intimidating enough that Anthropic really should have called this something else. Don't let the name put you off.

📢 Share The Signal

If The AI Optimist helps you think clearer, forward it to someone else navigating the shift.

Not quite landing? Hit reply. We read every message.

Hugo & Ben

• Startup Sherpas is the Social Enterprise that Hugo founded. You can find out more at www.startupsherpas.org. www.Startupsherpas.co.uk offers paid work experience for teenagers 13-19yrs old that parents can access directly for their teenagers and that funds our impact work where we’ve already supported over 5000 teenagers with paid work experience opportunities.