A colleague pointed me to this yesterday - https://gemini.google/overview/gemini-in-chrome/. (Of course there is so much happening at the moment that it is very hard to keep up).
As well as this we also had yesterday the announcement of Jony Ive joining OpenAI - video here - https://openai.com/sam-and-jony/.
We also had Google preview a diffusion model - read about that here - https://simonwillison.net/2025/May/21/gemini-diffusion/ (Important because these models allow effectively instantaneous responses, thus eliminating friction in their use).
These three pieces of news might seem somewhat unrelated, but I think what they collectively point towards is the industry moving towards ubiquitous availability of intelligence.
Matt Webb wrote about what that might look like back in 2023 - https://interconnected.org/home/2023/10/06/ubigpt
Somewhere there is a band that divides our problems between ones that we didn't think computers could help with at all - opening the possibility of entirely new ways of tackling those problems, and problems which are irreducibly complex, and where more "intelligence on tap" doesn't help because their solution requires social or institutional change. In the band between these two there will be much confusion, and the boundaries will be very fuzzy.
Large code bases will be the first of these domains to fall under the spotlight, and that will transform the area that I work in, but it will not be the most impactful or important area where these technologies will play a role.
Prequisets will be:
- Ubiquity of access to these tools.
- Response speed.
- Having folk be able to have a sense of the context window that the tool is giving them leverage over.
- Having that context window be simultaneously wide and accurate.
- Access into the biases that are operating.
As well as this we also had yesterday the announcement of Jony Ive joining OpenAI - video here - https://openai.com/sam-and-jony/.
We also had Google preview a diffusion model - read about that here - https://simonwillison.net/2025/May/21/gemini-diffusion/ (Important because these models allow effectively instantaneous responses, thus eliminating friction in their use).
These three pieces of news might seem somewhat unrelated, but I think what they collectively point towards is the industry moving towards ubiquitous availability of intelligence.
Matt Webb wrote about what that might look like back in 2023 - https://interconnected.org/home/2023/10/06/ubigpt
Somewhere there is a band that divides our problems between ones that we didn't think computers could help with at all - opening the possibility of entirely new ways of tackling those problems, and problems which are irreducibly complex, and where more "intelligence on tap" doesn't help because their solution requires social or institutional change. In the band between these two there will be much confusion, and the boundaries will be very fuzzy.
Large code bases will be the first of these domains to fall under the spotlight, and that will transform the area that I work in, but it will not be the most impactful or important area where these technologies will play a role.
Prequisets will be:
- Ubiquity of access to these tools.
- Response speed.
- Having folk be able to have a sense of the context window that the tool is giving them leverage over.
- Having that context window be simultaneously wide and accurate.
- Access into the biases that are operating.