The Goldfish Brain
Forgetting is a feature, not a bug.
AI coding assistants are brilliant goldfish. They solve complex problems in seconds, then forget everything you just told them.
The Real Skill in AI Coding Isn't Prompting
It's managing context so your brilliant-but-forgetful AI partner can actually keep up.
AI coding assistants are brilliant goldfish. They can solve complex problems in seconds, but forget everything you told them moments later. A recent discussion among developers revealed what many are learning the hard way: the bottleneck isn't prompt engineering, it's context management. Without a persistent, well-managed source of truth, your AI partner is always starting from scratch.
The real story is the shift from treating AI as a magic box to architecting its environment. Top developers are externalising the AI's 'brain' into the repository itself, using dedicated files like `AGENTS.md` to store architectural decisions, coding standards, and project goals. By creating atomic sessions for each task and frequently resetting the chat, they prevent 'context rot' and force the AI to rely on the documented truth, not a flawed short-term memory.
This isn't just a workaround for limited context windows; it's a new discipline. It makes AI collaboration more predictable, reliable, and powerful. Stop blaming the goldfish for its memory and start building it a better aquarium. Every developer using an AI pair programmer needs to master this.
Your New AI Teammates Are Here
While you're busy managing your coding assistant's memory, a new wave of specialised AI teammates just clocked in to handle the rest of the work.
Rumi.ai: The meeting assistant that actually remembers everything.
This turns chaotic meeting chatter into a searchable team brain, solving the collective amnesia that plagues every project. It ensures decisions and insights actually survive past the video call.
Super V2: Your company's brain, now with a better search function.
It tackles the same context problem as our main story, but for organisational knowledge. It’s about making institutional memory accessible on demand, so you can stop asking 'where did I see that?'.
Querri 2.0: The data analyst that speaks plain English.
This democratises data by removing the technical barrier to entry. It turns anyone into a data analyst, which is both incredibly powerful and slightly terrifying.
AI Is Building Its Own Tools Now
It's one thing for AI to help with your work, it's another when it starts building its own infrastructure.
InsForge: The backend builder for your AI coding assistant.
We're now building platforms specifically for AI agents to use, not just humans. This is the plumbing for the next generation of AI-native applications, and it's a huge tell for where the market is headed.
Baserow 2.0: An open-source Airtable with its own AI assistant.
Open-source, self-hostable AI tools are a direct challenge to the closed ecosystems. This gives developers complete control over their data and their AI-powered workflows, which is becoming increasingly important.
Quick hits
Typeless: Finally, voice dictation that doesn't make you sound like a robot.
It cleans up your 'ums' and 'ahs' to turn rambling thoughts into polished text, making your keyboard feel suddenly obsolete.
Grok 4.1: xAI's new model is here to feel your pain.
Boasting 'massive improvements in emotional intelligence,' this is the latest attempt to make conversational AI feel less like a machine and more like a confidant.
Tiny Mario: Super Mario Bros, but in your URL bar.
A brilliant, absurd proof that developers will turn literally anything into a gaming platform if you give them half a chance.
My takeaway
The real challenge of the AI era isn't building smarter models, but designing smarter ways to manage their context.
We keep chasing bigger context windows, assuming more data is the answer, but the real breakthrough comes from curation, not scale. The best developers are becoming librarians for their AI, meticulously cataloguing knowledge so the model can actually use it. This transforms the job from prompt engineer to context architect.
We are moving beyond simple chat interfaces to building persistent, externalised memories for our digital partners. It's about creating systems of intelligence, not just powerful chatbots. The next wave of innovation will be in how we structure and feed information to these systems.
What does your team's external AI brain look like?
Drop me a reply. Till next time, this is Louis, and you are reading Louis.log().