What did Sir Clive Sinclair actually change?

The price, rather than the specification. Before him a computer was somewhere you went — a university, a laboratory, a company that could sign off four figures without flinching. He had already done the same trick with the calculator, which was desk furniture with a mains lead until he put one in a pocket. Then the ZX80, the ZX81, and in 1982 the Spectrum, at a price an ordinary family could reach without a conversation about whether they could afford it. The 48K was modest even then. Getting the machine past the hall and up the stairs was the achievement.

Why load all your context at the start instead of fetching it when you need it?

Because my pool of possible context is small, bounded, and nearly all of it bears on nearly every session. Fetching on demand wins when the pool is huge and sparse and unpredictable, which is what a general tool serving lots of people doing unrelated jobs looks like. Loading the lot up front wins when it is one operator's own business. It is the cassette model rather than the disk model, and I did not choose it so much as inherit it.

Was the C5 a failure?

It sold badly and it was laughed at, which are two different things and only the first is a fact about the product. A small electric single-seater for short trips round a town is a description of a vehicle category you can now see from any window in Britain. He was roughly thirty years early, and the market prices early and wrong at exactly the same rate.

Does a big resident context cost more to run?

Less than the arithmetic suggests, because a byte-identical prefix is written once and then re-read at roughly a tenth of the input rate for the rest of the session. What decides the bill is how stable it is rather than how big. That doesn't make it free, and it doesn't stay true on its own either — a resident load that keeps growing will move past whatever you last measured, and nothing will tell you it has.

Everything I Thought I Invented, Sir Clive Sinclair Got There First

I spent a year building a way of working I was proud of. Then I went back and looked at where the ideas came from.

Every one that mattered was already sitting in a black plastic wedge with rubber keys, on a bedroom desk in 1982. I didn’t invent any of them. I inherited them from Sir Clive Sinclair, who spent the last forty years of his life being laughed at for a car.

The box

That kid had a machine with 48K of memory, and everything about the way I load context now came off it.

The 48K was never the achievement. The price was.

Before Sir Clive Sinclair a computer was somewhere you went — a university, a laboratory, a company that could sign off four figures without flinching. A calculator was a piece of desk furniture with a mains lead coming out of the back of it. He put the calculator in a pocket.

Then the ZX80, then the ZX81, and then in 1982 the Spectrum. The number that mattered about that machine was never the memory or the clock speed. It was that an ordinary family could buy one without sitting down first to discuss whether they could afford it.

He got a computer past the hall, up the stairs and onto a desk in a child’s bedroom. Everything I do for a living runs downstream of that. It took me until this year to notice.

Three things I was sure were mine

Load it all before you start

A ZX Spectrum loaded from cassette, and that load was the expensive event of the entire evening. You’d line the tape up and press play, and then wait several minutes while the border strobed and the machine screeched at you. At the end of that either the program was in memory, or you got R Tape Loading Error and started the whole thing again.

There was no disk to page from and no way to fetch a routine halfway through, so you paid the loading cost once, up front, on purpose, and everything the program did for the rest of its life it did out of RAM. If it wouldn’t fit you made it fit before you started, and you never made it fit by planning to go back to the tape.

That’s a complete cost model, and that kid had the whole of it long before he had a word for any of it.

Here’s what I do now, and I thought this one was mine. Nothing gets fetched mid-job.

Before I ask Monica anything at all she’s holding the kernel, the routing tables, every principle the business runs on, the platform traps for Astro and Netlify and Brevo, and the whole wiki rendered down into chunks she can take in one pass.

Then I say start client service, or load energy pages, and there’s no pause and no lookup and no let-me-just-check. The work starts in the first sentence, because the context got there before the instruction did.

The mainstream answer is the opposite and it has a proper name from the same era of computing: demand paging. Keep the resident set small. Fetch the page when the fault hits. Evict whatever has gone cold.

It’s a lovely piece of engineering and it’s been running quietly underneath everything you’ve ever used, and retrieval-augmented generation is that same idea with a semantic index bolted on where the page table used to sit. For most workloads it’s plainly the right call.

It is not the one I made, and I made the other one for the same reason a Spectrum did: the pool is small enough to hold, so hold it.

Forty years on the machine is different and the constraint is long gone, and I still load the whole world before I start.

The constraint is the design

Nothing about 48K was generous, and nobody working inside it thought of the limit as a feature. You got on with it.

What came out the other side was a generation of people who couldn’t afford a single wasted byte, and who therefore wrote as though every one of them was theirs. It was.

My kernel file — the one document every session runs on — has a hard cap in bytes and refuses, every time, to grow past it. When something genuinely needs to go in, something else comes out. I’ve defended that cap in writing more than once.

It’s a tape that only holds so much. I’ve rebuilt his constraint on a machine with no constraint at all, because the discipline turned out to be worth more than the headroom.

The machine expects you to make something

This is the one that took longest to see, and it’s the one that matters most.

A Spectrum booted to a flashing cursor. Not a menu, not a store, not a carousel of things somebody else had made for you to consume — a prompt, and the assumption that you had something you wanted to type into it.

The magazines printed listings on the page and you copied them in by hand, line by line. The discipline underneath the typing was that every line had to be present and agreeing before anything would run at all. You could load someone else’s game off a cassette. Or you could make the clock work.

I do this job through a terminal. There’s no dashboard, no drag-and-drop, no panel of somebody else’s buttons — a prompt, and the assumption that I have something to type into it.

When I tell a client the most useful thing I know, it comes out as own the ground you stand on and make something real on it, and I’ve said that sentence in rooms in Telford for 27 years believing it was mine.

It came off a machine that shipped with BASIC in the ROM and a cursor waiting.

Being early looks exactly like being wrong

He kept going after the Spectrum, and that’s the part people leave out.

A flat-screen pocket television, years before anybody wanted one. And then the C5 in 1985 — a small electric single-seater for short trips round a town — wheeled out into the kind of national laughter that follows a man for the rest of his life and some way past it. It’s a vehicle category you can now see from any window in Britain.

He wasn’t wrong about the C5. He was about thirty years early, and the market prices early and wrong at exactly the same rate. In 1986 he sold the computer business to Amstrad.

He died in 2021. A lot of people remember him for a vehicle almost nobody bought. I remember him for a cheap black wedge with a rainbow stripe that taught a generation to type.

He wasn’t trying to make computing precious. He was trying to make it ordinary, and cheap, and yours.

Sir Clive Sinclair would approve, and not for sentimental reasons

Everything he built was aimed at one thing: getting computing out of the institutions and into the hands of somebody ordinary, cheaply enough that she’d actually do something with it. Not a lab. Not a company with four figures to sign off. A person, at a desk, at home, making the thing work herself.

I run thirty covers out of a terminal, on my own, in a town he’d have recognised. That is not a clever new model of business. That is his model of computing, still doing exactly what he built it to do.

The machine got cheap so the person could get good. That was the whole point, and I spent a year thinking I’d come up with it.