Podcast

5 min read

Would you run your website on LinkedIn?

Season 2, Episode 2 of Beside Ourselves: Build or buy, the AI edition

No company would run its corporate website on LinkedIn. Yet plenty run all their AI on a single commercial LLM: one hyper-bundled package, owned by someone else and built for someone else’s goals. Behind the analogy sits IT’s oldest strategic question, build or buy. With AI, the answer hasn’t changed: buy what every business needs, build what makes you different, and don’t let the technology scare you off.

Download the free guide

In our Season 2 opener, and in our guide Below the Surface, we made a slightly flippant point: using a commercial LLM as your only AI system is a bit like running your corporate website on LinkedIn. This episode, we unpack it properly, stretch it as far as it will go (and then a little further), and get to the strategic decision underneath.

Why nobody hosts their website on LinkedIn

Giles started with the obvious: you’re renting, not owning. That has its pros, and it has its cons.

  • Different goals. LinkedIn is a Microsoft asset that exists to make money, from advertising and subscriptions, and not necessarily for you. You’re a secondary part of somebody else’s process.
  • It looks like everyone else’s. You get a different banner and logo, but every page comes out of the same box. You operate within their rules.
  • The data isn’t yours. You can see some of it. The bulk of the value, and the insight, stays with the platform.

In exchange, it’s easy, low-cost or no-cost, and it gives you a shop window. That’s a fair trade for a social presence. It’s not a trade any business makes for its own platform. And everything flows from that first point: it’s not your platform.

Five things we discussed

1. Strategy mostly comes down to build or buy

Theo’s view is that strategy, whatever you do, boils down to one decision: build or buy. In practice, it’s never one or the other. It’s a range. Buy what doesn’t make you unique — email, payroll, accounting, hosting — and treat it as a utility. Build what does, because that’s where you want control.

CRM is the classic in-between case. Infinitely configurable, and yet whichever one you choose (insert your favourite, or most infamous, here), you end up bending your process to fit the system.

2. Building comes with a long tail

We’ve both seen build go wrong. In the early 2000s, every web agency built its own content management system. When the relationship broke down, the client’s website was stuck on proprietary software nobody else could touch. That’s a big reason why so many firms switched to buying ready-made platforms: portability. (Giles used to run a CMS company, so he speaks with some feeling.)

Then there’s the “Jim used to do that” problem: a back-office system the business depends on, built by someone who left years ago. Building means key-person dependency, plus patching, security and upgrades for as long as the system runs. Hence the horror stories of government systems still running on Windows 95, because moving off them is too painful.

3. Buying has its own trap

The counterweight is vendor lock-in. At some point, you commit to an ecosystem — Azure, AWS, whoever — and getting out is hard. But Giles argued the risk is smaller than it used to be, because “build” no longer means a team of engineers writing everything from scratch. Today, it mostly means integrating and configuring existing components, connected through common standards. Much closer to plug and play.

4. Commercial LLMs are hyper-bundled

This is where the analogy earns its keep. ChatGPT, Claude and Gemini give you the model, the infrastructure, the interface, the data layer and the connectors in one box, just as LinkedIn gives you hosting, publishing tools and analytics. Most people never consider unbundling them. But you can choose the model, the hosting, the interface and the business logic separately, and there are open-source and commercial tools at every layer, so you’re not starting from scratch.

To be clear, we’re not saying commercial models are no good. They’re fantastic, and your own solution may well have one at its core. Set up properly, though, you can swap Claude out for ChatGPT, Gemini or whatever comes next when it suits the job better. And nobody’s suggesting you train your own model from the ground up. For most businesses, that’s neither affordable nor necessary.

5. Your data decides where your AI runs

Giles gave the example of an engineering firm whose whole business is its product, with lab, test and manufacturing data that amounts to its IP. Pushing that into a cloud-based model is a non-starter, and their AI policy (or their client’s) probably wouldn’t allow it anyway. So they take an open-source model, run it on their own infrastructure, and point it at their own data. They didn’t build the model, but they control the whole stack. The same logic applies to data residency, sovereignty, compliance and accreditation.

That’s where the traffic lights from our guide come in. Public information, like your website or your Companies House filings, can go through frontier models; they’ve probably consumed it already. Hypersensitive data, like IP or patient records, can still benefit from AI, but under strict control, and not in a frontier model. As Giles put it: if it’s a nail, it doesn’t need a sledgehammer.

Embed, don’t host

Before officially retiring the analogy, Theo broke it one last time. A hybrid AI architecture looks less like LinkedIn and more like YouTube. Nobody spins up a video server for a handful of clips. They upload to YouTube and embed the videos on their own site. In the same way, you design your own stack and bring in frontier models where you need their power, and where the data allows it.

Most businesses already have both a website and a LinkedIn page. Most will end up with a blend of AI tools too, sitting at different points on the spectrum depending on their data and the rules they operate under.

Don’t let “build” scare you off

Giles’s closing thought: the build-or-buy decision hasn’t really changed. Build what’s uniquely you; buy what every business does. AI only appears more complex, and that shouldn’t put anyone off. Both approaches are within reach for businesses of every size.

Theo’s closing thought came from what we’ve seen in the wild. AI adoption often stalls because a company concludes it can’t feed its data to a commercial system, and stops there: summarising, drafting emails, the usual superficial uses. The jump to building their own stack just isn’t obvious to many people. Yet even firms with modest budgets and small IT teams can go a lot further. There’s a whole tier of adoption still waiting to be exploited.

Coming up this season

We’ll be back in a couple of weeks, picking up another subject from the guide.

Download: Below the Surface

A deep dive into full-trust AI for SMEs. Real AI adoption needs sensitive data, and commercial LLMs can’t be fully trusted with it. Our free 14-page guide sets out what full-trust AI requires:

  • 6 models compared, beyond the frontier labs
  • 5 layers of an owned AI stack, explained
  • 4 steps to get you started

Download the free guide

Prefer to talk it through? Book 30 minutes with Giles or Theo — no agenda, no pitch, straight answers.

A deep dive into full-trust AI for SMEs.

Below the Surface: full-trust AI

Real AI adoption needs sensitive data, and commercial LLMs can't be trusted with it.

A full-trust AI stack is easier and more affordable than most think.

We compare 6 models, explain 5 layers and outline 4 steps to start.

Read the free guide
Giles Thurston and Theo Paraskevopoulos