--- title: "The AI Tool That I Wish Existed" description: "Envisioning an AI-first productivity tool that anticipates, connects, and surfaces ideas rather than requiring explicit organisation" type: note pubDate: 2022-09-26T00:00:00.000Z topics: ["ai", "tools", "productivity", "knowledge-management", "creativity"] canonicalUrl: "https://actuallymaybe.com/blog/ai-tool-wish-existed/" --- > Envisioning an AI-first productivity tool that anticipates, connects, and surfaces ideas rather than requiring explicit organisation There is an explosion of productivity-oriented ideas and systems. Browsing YouTube, Twitter, or Reddit, one can find nine different ways to organise your Notion workspace, how Roam Research will transform your note-taking, the best ways to learn from notes, and how you *should* listen to podcasts. Like any novel efflorescence, it has become the subject of memes about its excessive nature. Systems for system's sake. It's not that any of these tools are bad. It's that an excessive focus on *building a system* diverts attention away from the actual goal: doing cool stuff. But for all these existing tools, they are still remarkably "old-style web" in conception. Yes, there are novel UI advancements, or new ways of organising data. But they are still built from primitives a developer in 2015, 2005, or even 1995 would recognise. ## A Vague Outline of the Dream Tool The tool I would love to exist, one that might fall into this semi-mocked category, is built around capturing, synthesising, and making information useful for creation. But, critically, it would be as hands-off as possible. It would take an *AI-first* approach, integrating synthesis and usage directly into your current work habits. Existing tools would connect, existing applications could talk to each other, and your development tools would remain in place. All the while, this imagined tool quietly links everything together. ## A Brief Note on What "AI-First" Means What does an AI-first conception actually mean? Perhaps the best way to think about it is the difference between: - Flickr vs. Instagram - Google Maps on desktop vs. Google Maps on mobile The differing platforms change the product. They filter, reshape, and constrain the design. Trade-offs differ. Priorities differ. Entire feature sets evolve based on the platform that comes first. AI-first versus old-web-first is a similar divide. With AI-first, explicit user actions are minimised. Anticipation, prediction, and curation take centre stage. With old-web tooling, you are placed into a rigid frame of multi-layer menus, configuration options, and modular UI pieces. Of course, the two worlds overlap, just like desktop and mobile apps overlap. Some AI model configuration resembles setting up a Notion database. But this similarity is superficial. AI-first configuration has far-reaching effects that extend beyond the user's direct understanding, much like a fly-by-wire system intercepting and adjusting a pilot's control inputs in ways impossible for a human alone. ## So What Is the Tool I Think Is Possible? Current note-taking and knowledge-capture tools *assume* a human-directed model. You explicitly add or link content. You capture context. You build metadata. Inspiration may happen by chance, but the burden is almost entirely on you. ### Queues The tool I envision begins as a queue-management and ingestion engine, pulling in content from Twitter likes, Reddit upvotes, Google Docs, downloaded PDFs, catalogued, curated, spliced, and enhanced with contextual threads. ### Context That Twitter thread about new deep-learning research? I want the tool to: - save the PDF link - connect the discussion to that subreddit thread about fatigue in wind-turbine blades - remind me about both the next time I open my mechanical-stress Python library - proactively surface connections I didn't explicitly create ### Summarisation and Co-Pilot The tool should pull from my marked content in Pocket or Instapaper, plus the articles linked in that Twitter thread, and generate useful summaries. Then, when I'm writing in Google Docs or programming, it should know which ideas are worth resurfacing. God help me, but imagine a version of Clippy that *actually works*: > "I see you're writing a blog post on computer-vision research. What about that interesting paper you saved two months ago? The author explored something tangentially related. Here's how it might fit." ### Search An AI-first tool should allow vague but effective search across my content. Vector search already shows promise here. Even basic OCR on Twitter images would be a huge win. Let me search **"deep learning tesla model code"**, and have it surface a Reddit post I upvoted linking to a GitHub repo for a self-driving toy model. No, it's not literally the same thing, but the connection is obvious to a human. With a small dataset (my own interactions), plus contextual inference, the search becomes incredibly powerful without needing to comb the entire internet. ## Conclusion A tool for ingestion, curation, organisation, and search, sitting beside you as a creative copilot, connecting ideas you've encountered and surfacing them when relevant. This is all possible. And it could massively expand the creative potential of millions of people. More than that, such a tool would break down the useless ideology of siloed data. Creativity thrives on connection. You want this GPT-Creative companion enhancing *every* piece of content you encounter. Discover a new application? Grant access to your creative partner. Let it chew through the possibilities and throw back new connections, so you can focus on building things first, without worrying about systems or optimal organisation. A tool like this would not just store information. It would *unlock* it. ## Further reading - [Notes on Visual Interfaces, AI Tooling, Fast LLMs and Decision Models](https://actuallymaybe.com/blog/notes-visual-interfaces-ai-tooling-fast-llms-decision-models/) — Why agentic work is stuck in text, and what canvases, fast models, branching version graphs, and decision models might add. - [Emerging Field of Epistemic Engineering or Stochastic Management](https://actuallymaybe.com/blog/emerging-field-epistemic-engineering/) — Practices around context, uncertainty, and evals are hardening into a discipline, with belief states at its centre: probabilistic, grounded views of a work process. - [Freeport: A Lightweight Port Management Tool for Development](https://actuallymaybe.com/blog/freeport-lightweight-port-management/) — A thin wrapper around native commands to simplify port checking and process management during local development. - [Thoughts on Context and Conversation for Ideas](https://actuallymaybe.com/blog/context-conversation-ideas/) — Exploring how conversation and exploratory browsing create a 'mental latent space' where ideas emerge from unexpected connections - [The Different Problems of Searching Global and Local Context](https://actuallymaybe.com/blog/global-vs-local-context-search/) — Exploring why global-context search is easier than local-context search, and what tools built with local context in mind might look like