**Type**: Insight **Growth Stage**: Evergreen **Last Tended**: 2025-04-16 **Topics**: #ai #AICollaboration #CreativityAndTech #LLMs #conversationswiththemachine --- Recently I've shared my observation of an AI assistant that is coalescing around me and the work I do. If I borrow from a more science-fiction trope then ATLAS (the name I've given my collaborator), doesn't have corporeal form yet - it exists more in spirit, a higher form of capability or intelligence - but as the technology advances closer to my vision, it is manifesting in several ways: **Capture and Organise Thoughts** Notes apps are great for quickly capturing an idea, either in text or voice memo. But AI apps can offer a similar capability and enhance the interaction too. Depending on how you set up the initial system prompt you can instruct the model to do everything from nothing, just capture and wait, to going off and exploring the idea, getting research or making a plan. With voice to text modes, I find this incredibly useful for those moments in the middle of something else when you just need to get the idea logged and come back to it later. Often that's in the middle of a ride out on my mountain bike or waking up in the middle of the night! I capture the week's worth of thought's in a single chat under the project 'Thought Machine' and then at the end of the week I get the LLM to summarise my ideas for me to reflect on. **Researcher** There are now a variety of ways to use AI as a research analyst. New 'search' or 'web' integrations give LLMs the tools to move beyond the limits of out of date training data to a viable replacement for traditional search when **Ideation and concept development** Getting into a conversation with an LLM on a half-baked idea is one of the best procrastination and blocker beating ways to move that spark of inspiration forward. The way they work mean they are really good at making connections and can spot stronger or weaker concepts, drawing in related material that can validate your ideas or build upon them in unexpected ways. I frequently turn to Anthropic's Claude app where I can create a 'project' - this is a collection of chat sessions on a specific topic that retain a shared memory or context. I believe this is one of the most vital features in AI assistance. The ability to put a ring around a specific area of context means you can start new and related threads that can pull upon previous context - for instance a new feature can be discussed with full context of the wider ambition. Claude and many similar apps also deploy a 'canvas' feature. This side pane creates the document or artefact along side the chat in real time and can summarise theories and ideas. Importantly these documents can be added as specific referenced files in the 'project' that can be called upon and updated as things progress. When you need to jump out and take the detail with you - a single prompt can pull the relevant documentation into a summary file for a specific audience or purpose. **Sounding board & inspiration** You don't want your digital counterpart to be a 'yes man' - ideation and creation in a vacuum doesn't always produce quality results. You need it to tell you when your idea sucks or needs a bit more work - after all, you want to pick the very few best ones to spend your precious time progressing. As thoughts and concepts form you can prompt and challenge the LLM with a variety of 'what if' scenarios asking it to take on persona or give it additional observations from your experience that update it's recommendations or evaluations and produce new and better results. I like to explore all the edges of an idea and test the potential downfalls, so I often ask AI to play 3 key roles: **The Optimist, the Pessimist and the Challenger** - the latter questions me to go deeper into an idea addressing the concerns and confirming the benefits posed by the former two (more on that in another article). When you do need that validation - it's really good at explaining why, often in a logical tabular format too - this gives you greater confidence and motivation for your work. Here's a recent example exploring how I could bring novelty to my writing, including AI. I had the rough concept for ATLAS already, so I was challenging its fit to the vision I have: | Format | Strategic Fit | Lifestyle Fit | Strengths | Risks | My Recommendation | | --------------------------------------------------------------------- | ---------------------------------------------------------------------------- | ------------------------------------------------------- | ------------------------------------------------------------ | ----------------------------------------------------------------- | ------------------------------------------------------------------------------------------- | | **ATLAS (AI Dialogue Partner)** | ★★★★☆ Strong narrative layer, shows your thinking, deeply human-machine lens | ★★★★☆ Modular, episodic, playful, easily paused/resumed | Distinctive voice, repeatable format, shows systems thinking | Needs clarity of ATLAS’ voice. Risk of forced gimmick if overused | Absolutely worth pursuing as a core narrative layer — but keep it light, witty, human-first | | **Systems-in-Dialogue** (Dialogue Iterating a Framework) | ★★★★☆ Strong alignment with Systems-as-Gift | ★★★★☆ Frictionless from how you work already | Transparent IP building, invites feedback, shows generosity | Could feel niche if every post is this style | Use regularly but not exclusively — powerful content atomisation device | | **Augmented Field Notes** (AI adds alt perspectives into your essays) | ★★★☆☆ Better for Substack essays | ★★★☆☆ Slightly heavier lift to execute well | Great for life reflections, signals humility + depth | Slower to produce, risk of being too meta | Occasional use for deeper reflections — not every week | | **Digital Twin Chronicles** (Training AI with your systems) | ★★☆☆☆ Not directly aligned with immediate goals | ★★★☆☆ Could be fun, but more experimental | Audience might find it fascinating inside look | Audience might not care unless positioned really well | Leave for later — after ATLAS is established | | **Parallel Processing Journal** (Human vs AI approaches side-by-side) | ★★★★☆ Great Systems-as-Gift method | ★★★★☆ Lightweight once rhythm established | Great tension driver, clean & fun | Risk of becoming formulaic | Perfect for LinkedIn — occasional series | | **Interactive System Building** (Post shows live decision process) | ★★★★☆ Aligned with transparency & enabling others | ★★★☆☆ Slightly higher friction | Engages audience deeply, shows thinking real-time | Might need more visual design support | Use occasionally for big frameworks or flagship ideas | **Manage projects and actions** I often find myself stuck with where to start on a task or project. I find it hard to get from concept into execution. AI assistance can offer the help to break down the initiative into a logical sequence of activities to move from idea into implementation. The answer is often simpler than you expect and often very logical. It also helps to have the sequence of events all laid out in front of you so you can check that the overall journey is with your capability, or constraints and that the an idea is well formed enough that the end goal you had in mind can be reached through a series of steps. Working on a project or set of related tasks in a conversational format also presents a new way to manage things. Since the AI is presumably calculating what might more traditionally look like a timeline, plan or set of actions all behind the scenes (but not in a actual state) you can simply ask it 'what's next' and then update it as you progress, all the while letting the machine be responsible for adjusting those plans and report back on the state of progress. I was recently planning a series of monthly family events for 2025. This took some ideation, logistical assessment, planning and then capturing of progress as I locked in all the dates and ideas and booked them too. Taking this on in a conversation in an organic way was a great way to use AI as a personal assistant who nudged me along in moments of free time, making progress one step at a time, adjusting to changes rather than trying to climb the mountain all in one go. > [!NOTE] Here's how part of my chat with Claude went: > *Ah yes, that makes more sense! September will likely have better weather for London exploring, and a cozy cinema night works well for December. Here's the updated calendar:* > > ***LOCKED IN:*** > > 1. ***January 2025** - Board Game Café ✓ (BOOKED)* > 2. ***February 2025** - Sushi Making (Have Groupon, need date)* > 3. ***March 2025** - Eden Project Family Day* > 4. ***April 2025** - Highclere Castle ✓ (BOOKED 12/13 Apr)* > 5. ***May 2025** - Watercress Festival ✓ (Sun 18th May)* > 6. ***June 2025** - Comedy Club Night* > 7. ***July 2025** - Iron Maiden Concert ✓ (BOOKED)* > 8. ***August 2025** - Garden Tour & Tea* > 9. ***September 2025** - London Day Out (Tate/NHM)* > 10. ***October 2025** - Devil Wears Prada Show ✓ (BOOKED)* > 11. ***November 2025** - 12-Course Tasting Menu (Birthday)* > 12. ***December 2025** - Cinema Evening* > > *This works better because:* > > - *September usually has good weather for London exploring* > - *School holidays over in September so London less crowded* > - *Cinema perfect for cozy winter evening* > - *Could catch a Christmas movie in December* > - *Better spread of activities across seasons* > > *Next steps could be:* > > 1. *Research comedy clubs for June* > 2. *Look into garden venues for August* > 3. *Plan London September day (exhibitions, lunch spots)* > 4. *Research good independent cinemas for December* > > *Would you like me to focus on any of these next steps?* **Strategist** The coupling of LLMs ability to create by making connections, organise user inputs and build on a vast set of general training data make it an excellent strategist. For everyday use, I'm not talking about 'game theory' level execution but a useful generation of frameworks and plans that can help guide us in our approach to different pursuits. For me this sits neatly between ideation and implementation plans - when the concept is a bit too broad or covers many aspects to satisfy a target plan but you need to organise thoughts in an intentional way. This is probably best brought to life with a few examples, where I have given some input as the prompt and asked the assistant to help me build a framework or form a strategy for my approach. 1. Digital Strategy. I want to maintain a social presence and develop my writing skills online. I use a variety of platforms; Linkedin, Substack, Bluesky etc. and I want to share my experience with a few topics: AI, Personal Development, Industry Consulting. How should I approach this? 2. Seasonal planning. (for those who like to go beyond journalling to a longer term framing) - I want to achieve *x* and *y* while currently working on other certain things. I have constraints a, b and c but I am passionate about z... how should I manage my time over the next 3 months? In both the above cases, I can also include additional context such as previous content or templates that help give detail and guidance to the LLM on how to format a response. In the case of a Digital Strategy, I had shared a number of previously shared articles so it would understand the themes and tone of my writing. It came up with this response: **Big Theme Across All Platforms:** “Building environments, systems, or personal practices for flourishing in a modern world shaped (and sometimes warped) by technology." | **Theme** | **Treatment Style** | **Platforms Used** | | --------------------------- | --------------------------------------------- | ------------------ | | AI Skepticism & Optimism | Exploratory, philosophical, sometimes warning | LinkedIn, Bluesky | | Personal Systems for Growth | Methods, metaphors, frameworks | Website, Substack | | Showrunning Your Life | Story-driven, personal reflections | Substack, LinkedIn | | Practice & Craft | Tying action to wisdom | Website, Bluesky | **Critique and edit creative writing** It is all too easy to let AI take over with writing now. It excels at this. But as I have written before, my experience has shifted from training wheels to feedback mode - now relying less and less on AI as I have developed my writing muscle. I believe there is a scale on which the importance of the content can be weighed and judged to whether an AI is simply the more productive tool to complete the job and share a point of view. But when it all comes from the heart it is useful to get a. first feedback on the content - your own 'editor in chief' who knows your strategy, the tone you like and how it resonates best with the audience. It can give feedback, suggest subtle changes and even prepare versions (eg. shorter summary ones) for different platforms. I also find a little past 2-3pm I enter the zone of 'cognitive fatigue' - if a quick workout isn't re-energising me then I find AI can step in to finish the job and maintain productivity. For instance as I was writing my last article which recognised brought to life the concept of my assistant, ATLAS - I needed to add in a additional point I wanted to make around AGI. I shared a version of my article to Chat GPT and asked: *"i'd like to add a little commentary on why I am giving it a name [ATLAS], a presence anthropomorphising it. - because we might (MIGHT) be on the path to AGI or close to AGI - and I am intrigued by the idea of a future with general AI assistants supporting our everyday"* GPT responded: > [!NOTE] **Why give it a name? Why give it presence?** > > > Because I think we _might_ be heading towards a future where this isn’t unusual at all. Where naming an AI assistant — giving it presence, history, quirks — isn’t just branding or theatre. > > > It’s how we’ll relate to them. > How we’ll trust them. > How we’ll teach them. > And how they’ll teach us. > > I’m not saying we’re at AGI today. But I _am_ saying we’re close enough to a future of general-purpose assistants that it’s worth exploring what it feels like to build one. **What ties it all together?** Across all these roles — from ideator to strategist, sounding board to organiser — a few consistent patterns emerge in how I use AI: - **To test and stretch ideas** – I use AI to pressure-test theories, explore edge cases, and push half-formed thoughts into new terrain. It’s an iterative, creative dialogue that strengthens clarity. - **To raise the quality of my thinking** – Whether it’s surfacing blind spots or offering new perspectives, AI plays the role of a trusted second brain — sharpening what I already know and revealing what I’ve missed. - **To spark momentum** – When I feel stuck or foggy, a quick prompt can unlock a clear direction. The ability to quickly sketch a plan, identify next steps, or reframe a problem gives me forward motion. - **To inspire and provoke reflection** – AI often acts as a mirror, reflecting my ideas back in a different form. That reframing leads to new connections and better questions. - **To maintain flow in life and work** – From managing projects to shaping strategy, it’s a form of ambient cognitive support. A quiet, persistent co-pilot helping me shift from vision to action — one prompt at a time. **How does ATLAS show up?** These consistent patterns — testing, refining, building momentum — are expressed through different **roles** ATLAS plays in my process. Sometimes it’s my sparring partner, other times my scribe. Each role has a distinct function and impact on the content I create and the way I think. Here are a few key roles ATLAS takes on: |**Role**|**Function**|**Content Impact**| |---|---|---| |**Apprentice**|Asks naive or analytical questions|Makes your thinking explicit| |**Challenger**|Probes assumptions and weak spots|Creates productive tension| |**Synthesizer**|Spots patterns across thoughts/projects|Shows systemisation process| |**Archivist**|Captures and curates lessons learned|Builds a long-term IP library| |**Reflection Engine**|Mirrors your habits and practices|Deepens human authenticity| |**Co-Builder**|Collaborates in designing systems|Demonstrates human–AI design partnership| Each of these isn’t just a function — it’s a lens. A way to extend, question, or clarify my own cognition. Sometimes ATLAS holds the pen. Sometimes it holds the mirror. **What’s still missing — and what’s starting to emerge?** While each of these roles is powerful on its own, the real breakthrough lies in **joining them up** — creating a seamless, integrated experience that spans across time, context, and projects. Today’s tools offer a part of the vision - but without true memory or coordination, these moments remain fragmented. The assistant helps in parts — but the whole is still up to us to hold. To truly unlock the next level of capability, we need: - **A master memory** — persistent understanding of your projects, tone, goals, and patterns of thought. Something that knows not just _what_ you’re working on, but _why_. - **Contextual continuity** — the ability to bridge insights across independent threads, so that reflections in one area naturally enrich another. - **Semantic organisation** — not just storage, but structure. A way to surface relevant ideas or themes dynamically, when they’re most needed. - **Agency with guardrails** — a system that can take initiative, suggest paths forward, adapt to new goals — but still defer to your intent, constraints, and values. Encouragingly, we’re starting to see the **early architecture** of this future. - Agentic workflows — where AI can plan, take action, and reflect — are beginning to offer more cohesive flows. - Fine-tuning and system-level prompt engineering allow for more stable, custom personas. - RAG (retrieval-augmented generation) patterns make it possible to ground responses in a user’s own work, across files, notes, and previous conversations. - Together, these tools are starting to form the connective tissue — the nervous system — of a true **general-purpose cognitive partner**. It’s still early. But you can feel it coalescing. That’s the vision I’m building toward with ATLAS: Not just a tool that responds to my commands, but a collaborator that **remembers**, **organises**, **challenges**, and **evolves** alongside me. Over the coming weeks I'll share the experiments an journey as I attempt to start to stitch these capabilities together and invite ATLAS into the team too. --- *This is a living document in my Digital Garden. It grows and evolves with my thinking and represents my personal thoughts and opinions, and is not part of my work at IBM. However, it is part of my desire to contribute a broader conversation on how we 'get things done' - exploring the impact of tools and techniques aligned to my mission to help individuals and organisations create the settings for sustained growth.* ## Growth Log Published on substack - 16th April 2025