From bookmarks to embeddings: building your Leadership Library

Reading time: 7 mins

For a long time now, one of my key practices as a manager and leader has been building out my own personal bank of resources; my “Leadership Library” if you will. This has been a practice that has proved invaluable to me, where the importance has only compounded over time as the Library has grown. Now, in the age of AI-all-the-things, I’m starting to explore expanding it into the world of agentic capabilities.

Just do it

If there’s one takeaway I’d like you to have from this post, even if I’ve already nearly lost you at “AI”, it’s that I think everyone should keep their own stash of resources. Some of the things this has helped me with includes:

Finding relevant references quickly, whether it’s a bookmarked link, or your own template for running a particular process (think things like performance calibration, running 1:1s, or keeping brag docs)

Reminding yourself. Despite my best intentions, my long term brain storage often feels like it’s mostly full of song lyrics and niche Japanese vocab, and I’ll find myself forgetting details about topics I once situationally knew in great depth. Sometimes when I’m looking through my resources or blog posts I’ll re-read something from years ago and will pleasantly surprise myself with how smart past-me sounded, and absorb it all over again.

Having an easy source for sharing with others, whether that’s on your public blog for others to use, or a more private one for you to wheel out when necessary. Making a persistent bank for your own reference avoids me reinventing the wheel, re-writing lengthy messages on common topics like “what it means to be Senior”, or “how to give and receive feedback”. I tend to apply the usual rule of automation here – if I find I’ve had the same conversation more than a couple of times, it’s usually helpful for me to create a resource or article about it.

Since working independently, jumping into new environments, and working with coaching clients more frequently I’ve been especially grateful to myself for this existing store of wisdom. It’s made it a lot quicker and easier to follow up conversations with a set of relevant additional reading materials or practical guidance to pass on.

Resources can be anything that helps you

My ultimate goal is always to publish things on my blog so that they can hopefully help people more broadly (👋🏻), and over the years I’ve done so with posts like the following:

However my blogging has suffered since becoming a parent, and there’s a lot that’s never made it into a ‘proper’ public post. I have everything from bookmarks through to scrappy bullet lists and tweakable/repeatable resources in my more private bank of references. I customise and reuse these frequently in my work, including topics like:

  • Effective self reviews during performance review season (tweaking it to reflect what each company values)
  • Feedback (perhaps making reference to the specific tooling used in-house)
  • Activities/workshops that I’ve run
  • Career conversations
  • (and more)

In my last role at Bumble I gradually added to and shared out a whole folder of my notes, centralising anything and everything that I was providing to teammates, and at the same time giving access to the wider engineering community in case it was helpful more broadly.

Evolving for an agentic world

Of course, things are rapidly changing in this world. Where in the past I’d relied on folders, manually created documents, indexes, and a decentish search, we now have our new agentic tools to factor in.

Recently my ex-colleague and all round very smart person Ola Sendecka introduced me to Nate B Jones’ take on an ‘open brain’, which caught my eye. The idea is to create and own your own portable repository of thoughts, independent of any vendor lock-in. You can add to and semantically query the knowledge base, and crucially using MCP it allows you to connect to various AI clients or tools like Slack, allowing you to add to, update, or retrieve from one single source of truth whatever the tool you’re working in, saving you having to repeat context in different places.

This sounded like a great partner for my blog and longer-form articles. Now, rather than adding in isolated links and snippets of thoughts that need bringing together later, I can add them in on the fly, create an embedding, and then retrieve them as and when in different contexts. The system also automates identifying themes and entities like people or places, so I can ask a tool to retrieve every reference I’ve been given by a certain person (maybe I know it was from them but I couldn’t remember exactly what it was), or pull out all of my notes on one theme from all of the conferences I’ve ever been to.

It’s still early days for me, and given that I’m still on maternity leave I’m not yet using it for things like 1:1 notes, tasks, or weekly reviews, and I haven’t yet integrated with tools like Slack or my calendar. However in the future I can see it being very valuable for tying everything together, and hopefully helping me to stay on top of work with less time wasted on admin.

Words of caution

…But as with any good technology post (especially ones involving AI), there’s of course the usual caveats and plea for responsibility.

Very early on, I realised that I had some crucial design decisions to make. I could switch this on and rapidly start capturing absolutely everything that goes into my AI assistants, from questions about how best to support the newts living in my pond, through to Japanese grammar points that haven’t stuck, or ideas for kids activities that are fun for me too. Now, as I own this data my hesitation was less about privacy, but I wasn’t yet sure if I wanted to go all-in on a true representation of everything that I’m thinking about on a daily basis. I started off very deliberately locked down, requiring the AI tools to ask explicit permission to write anything to the database, although this proved to be pretty tedious getting a prompt for every embedding it wanted to create. More likely I will slowly expand out and update the guardrails to contain the scope, although this is obviously fallible.

Additionally, your choice of tooling doesn’t necessarily entirely remove the use of third parties, and that may come with issues of cost, privacy, security, or black-boxiness. Nate Jones has two different approaches on his blog, one involving Slack/Notion/Zapier/Claude or ChatGPT, and one using online database hosting, an edge function, and an MCP server. For your use case you may wish to run everything entirely locally – for example I spoke to Ola about her idea to create a local version using Python, Qwen for a local model, SQLite, and tying in Obsidian for a visual interface.

Running locally may solve some issues but create other questions: for example if you’re doing it to work on a company laptop, what happens when you leave that company? Can you safely export the context and take it with you? Would you be able to safely and confidently sanitise anything sensitive? Or will you be stuck feeling a huge loss if you have to leave everything behind? Is there a way you could safely edit and migrate the most useful knowledge to your successor? These are all questions that come with a lot of uncertainty and unknowns, but are worth trying to ask yourself (and your company!) at the outset. Ensuring that you credit appropriately, that you aren’t unlawfully capturing anything that infringes on company confidentiality or IP, and that your LLM(s) of choice respects the guardrails are all super important.

AI won’t be replacing my full Leadership Library just yet

As for me though, I’m looking forward to AI’s help with the more scrappy pieces of brain dumping, and seeing how it can help me with looking for patterns and opportunities, but I’m not expecting it to replace everything. I’m still intending to keep extending my huge bank of personal resources and individual content, and maybe even adding a human-written blog post or two along the way when I get a spare couple of hours 😉

Thanks to Janko Ferlič on Unsplash for the header image