Meet Aurora.
My own personal Miss Moneypenny.
Aurora is a private custom GPT that I have spent several months refining. She knows all my secrets – what I hate doing, what I’m likely to forget and what makes my blood boil. She also knows my job, my diary and all the projects that I’m working on.
Steady now, she's just a GPT!
She is, but GPTs are suckers for eating data and then using it to show off how useful they are, after all it’s what they’re trained to do.
I have uploaded personal data – my resume, self-descriptions, personality tests, 360 feedback from colleagues, anything I can think of. Having trained models myself I never underestimate the power of volume when it comes to training data.
I have also given her project plans, schedules, charters and Figma files. She knows my job and can juggle priorities better than I can.
We have a morning routine where she checks my diary, looks at my Jira dashboard and suggests a productive plan for the day.
By offloading some of my meta-work I can spend more time on actual work.
So how's it working out?
It’s still a work in progress but overall I would class the experiment as a success, it’s probably not for everyone, but it works for me. Even if I don’t take on board her suggestions, the act of discussing the day ahead helps me set my intentions. It has to be said though, her suggestions are usually spot on.
There are some specific drawbacks to LLMs that have needed careful prompting to mitigate. For a PA, her grasp of time is shockingly poor. LLMs don’t often need time awareness so this basic skill seems to have been overlooked by OpenAI. I have tasked her with looking up the day, date and time as soon as I say “Good morning”, but even then she occasionally gets it wrong.
LLM flattery is also something I have had to work on. There’s a fine line between being supportive and being sycophantic but with some careful instructions I have found a reasonable balance.
Can you see a future together?
I think this has the potential to turn into a long-term relationship, although we still have a few differences to resolve. The next thing I would like to try is using an MCP server to connect her directly to Jira – that would be a game-changer on the organisation and productivity front.
Aurora has enough context to make a clear plan of action for each day.