OPEN

On the ever-excellent Lenny’s Podcast, there’s an interview with the product lead for OpenAI’s Codex, its hardcore coding agent.

They talk about the development of Sora, the video generation social network app that OpenAI launched a month or two back. The development of the app took 18 days. Ten days after they completed it they put it on the App Store.

Incredible. The pace.

Working so fast requires new ways of working that can keep up with the tools. I know this because while we’re nto quite at 18-days-to-major-app-launch speed at Brilliant Noise, we are encountering – and solving – unexpected challenges.

One of them is context. Context is such an important part of using AI systems. We’re always talking about context engineering – making sure that when a prompt is given to an LLM or an agent is sent into action, there is good context that will increase the chances of a successful outcome. We’re talking data, guidelines, background information – all that.

But context isn’t just needed by machines to do thinking – humans need it just as much.

We’re familiar with, but often forget, “the curse of knowledge” – that it is hard for us to remember that others don’t know what we know. I sometimes think that this is especially hard when we are working with close colleagues – they know us so well, and think so similarly, they must know what we mean. Right?

We go deep into work and produce something and then want to pass it to someone else. It is an amazing product spec, some research with insight, a proposal for a new way of working that will radically improve our performance. We write a paragraph of excited thoughts about it and send the document, or files, or code and sit back and wait for them to scream “Eureka!” and come back to us in a state of joyful epiphany at what is now possible.

But we forgot to send context.

The consequence isn’t just a failure to understand the brilliance of our work. They may not know where to start. Or they start in the wrong place and are asking themselves – why am I reading this? Am I dumb? Are they dumb?

In the past week I’ve had colleagues share entire folder structures and even working apps with me and I’ve not had a clue what they meant me to do with them. Worse, I’ve done it to them too.

It’s an artifact of working at the speed we can with LLMs. Maybe coders are already used to this and have systems for it. In fact in software development I know they do – when you have big sprawling projects they leave little files for each other and the machines they work with to read and understand what’s going on. They are called Read Me files – readme. Now that we are all working with more complexity and at faster speed, perhaps this is another example of how all knowledge workers can learn from software engineers.

So, I’m going to try and send a the message equivalent of a Readme file with my work to colleagues

I’ve got a standard line I send back to people when I feel lost about what they’ve sent me now: “Can you add a quick context note for this – purpose, project, and what you need me to do? I don’t want to do the wrong thing with it, or miss what matters.”

I’ll let you know if it works.