Seconds out. Mythos v Spud

A few huge rumours doing the rounds at the moment with enough credibility we should be paying attention to them

Mythos first, because Mythos is the one that could change the shape of the year. The chatter is that it's a model so capable it can read enormous codebases end to end and find security holes nobody knew were there. The story doing the rounds is that it can't ship until the world's software has had time to patch what Mythos would otherwise expose. That part could be hype. What's harder to dismiss is that there are credible references to specific security and software vendors already having access to it, quietly using it to close vulnerabilities before the model is more widely available. Picture a horse helping to bolt its own stable door shut before it gets loose and causes carnage. That's roughly the situation. Any AI that powerful is a massive step forward from what we have seen so far. Remember when chat gpt 3 could only do a short essay and where we are now, this looks like a bigger leap.

While we wait for Mythos, Anthropic is rumoured to be launching Opus 4.7 this week. Not quite as seismic. Still a meaningful step forward, and probably the model most people will actually be using to show demos at Cannes.

Then there's Spud. OpenAI's latest. Not quite the same power hype as Mythos but another rumoured massive step forward over where we are today and likely to be in the hands of the public much sooner. Anthropic recently went passed OpenAI in valuation so they need this to be good. Spud v Mythos, one of those names is great. The other you can mash.

I don't know which rumour fully proves out and which gets quietly revised down. What I do know is that the rate of improvement over the last six months has stopped looking like progress and started looking like acceleration. Something bigger is coming. Probably several somethings. Probably soon.

Worth pausing on what happened with coding, because it tells you what's about to happen everywhere else.

A year ago, agents were the thing everyone was meant to be excited about. Useful. Not ready. The promises ran ahead of the product, as they tend to. Coding was the one place you could measure the gap honestly, because code runs or it doesn't. By autumn, almost all the code being written at any serious software shop was written by an agent and reviewed by a human. Not long ago programmers themselves were saying AI wouldn't disrupt them quickly. They were wrong. The ones who adapted fastest didn't disappear. They got more valuable. They stopped typing the code and started designing what the code does, at a scale and a speed that wasn't possible when they were typing every line themselves.

Coding went first because success is binary. The model writes a function, you run it, you know. Everywhere else, in media planning, in customer service, in legal review, in financial analysis, success is fuzzier and the feedback loop is longer. Adoption takes another six months or so to catch up. We're in that six months now.

When the next real leap lands, and based on the noise it's landing soon, the businesses that benefit will be the ones whose work is already written down in a form a model can use. Not locked in heads. Not living in the muscle memory of the senior who's been doing it for fifteen years and couldn't tell you how if you asked. Codified.

That's the prep work. Get the method out of the people. Put it somewhere a model can read it, run it, improve it.

For our world that means the actual craft. How you plan media. How you optimise. How search works, how social works, how measurement and attribution fit together when nobody's pretending. None of that lives inside a foundation model. It has to be encoded by people who've done the work for years and know what good actually looks like.

Incidentally, that's what TAU does. It's why we built the company the way we built it. Industry operators, including a few genuine legends like Zuze Gierlinska and Paul Cooper, sat next to AI developers writing the code. The operator knows what good looks like. The developer makes the model do it. Neither half works on its own. Together they turn craft that used to live in heads into something a new model can plug into the day it lands. So when Mythos or Spud or Opus 4.7 turns up, the foundations are already there.

The model is the leap forward. Your foundations are whether you can stand on it when it arrives.

Or as the great philosopher Rocky Balboa put it. That's how winning is done.