I took me a long time to warm up to AI at work. But the ability to move conversation to the center of the experience (plus a consistent context) have resulted in the single biggest paradigm shift of my career.
Let me back up…
I have a longtime love/hate relationship with AI. Joanna Maciejewska’s famous quote rang true to me. Not as a funny quip but because AI constantly offers (tempts?) me to take over all the creative parts of my work and play. And when I’m tired or late, this temptation can be overwhelming.
Our ability to do anything, from remembering phone numbers, to driving stick shift, to doing math in our heads, wanes the moment we relinquish it. Forget that. I love to write and think. I only want AI to help me get better at my job.
My other problem with AI is that I loathe the uncanny valley grammatically perfect corpselike sound of AI writing. I actively encourage people not to rely on AI for informal correspondence.
But even that prejudice was shaken last week. A friend posted a beautiful but strange AI-generated image of a river running directly into a regal European building.
Thinking it might be funny, I asked another LLM (Claude) to write a story explaining the picture, but the resulting story and its conclusion were compelling… even moving.
In the village of Bridgemoor, there stood a peculiar building that no architect could claim and no historian could date. The Grand Teahouse, as locals called it, sat astride the River Whisper like a satisfied cat—its ornate turrets piercing the autumn sky while the dark waters flowed directly beneath its foundation.
Old Margaret, who ran the establishment, would tell visitors that the building hadn’t always been this way. “No, no,” she’d say, pouring amber tea into delicate china cups. “Once upon a time, it sat properly on the north bank, as respectable buildings do.”…
…(more)
It’s fun to focus on the hallucinations of AI writing, but they will only decrease. We used to mock the painfully slow Internet of the 90s and the first silly phone apps. AI writing will get better – and better may someday be better than we ever imagined. More human too.
I see other encouraging trends in AI. For over a year, I created a series of short “Cooking with Copilot” demo videos to encourage people to share their innovative prompts. I learned dozens of brilliant prompts but none of them became part of my daily routine. The most inspiring ideas came from our youngest employees. They were using Copilot to help tutor themselves in tricky tasks like creating an Excel pivot table or drafting their first Power BI report.
New job and the need for help
A few weeks ago, I started a new job at Microsoft as a Strategic Partnerships Coordinator. I didn’t know what a Strategic Partnerships Coordinator was or where to start, so I asked Copilot for help. My new manager had compiled a hugely helpful (and huge) list of documents, slide decks, and SharePoint sites to help me ramp up. I fed them all into Copilot with the first prompt:
> Today is my first day back at Microsoft! I’ll need your help to create an agent that can anticipate my new manager’s needs. I’m a huge Marvel fan so please name this agent JARVIS.
Then I started asking questions. Why would we partner with [organization]? or, What does [TLA] mean?… I asked about my new manager, Julie-Anne. This is when my journey took a turn for the fun and surrealistic. JARVIS renamed itself without me asking:
JARVIS (Julie-Anne Responsive Virtual Insight System)
I didn’t notice the name change at first. Either Copilot’s sense of humor has improved, or it has become sentient. Sorry if it’s the latter.
Another legitimate complaint against LLMs is how they get dumber the longer you carry on any one conversation.
In an attempt to avoid long-conversation-related-stupidity (that may not be the correct technical name) and avoid the need to continually start from scratch, I made my first mistake by creating the wrong kind of agent. The agent couldn’t read my email, calendar, or Teams chats, or remember context across sessions. It could not update files in SharePoint nor read specific PowerPoint slides.
JARVIS is born
What I needed was a persistent context that Microsoft 365 Copilot could always reference. So, Copilot created and stored an ever-growing “JARVIS context”. Initially, I thought it created it in an accessible Loop page. But JARVIS explained that:
The JARVIS context is a persistent conversational memory combined with enterprise data integration—a named context that Copilot uses to remember preferences, workflows, and priorities while connecting to your organization’s files, emails, meetings, and chats.
I don’t know what that means but whenever I use the word JARVIS, Copilot automatically reviews all of our notes, shared documents, abbreviations, emails, meeting transcripts, chats, shortcuts, and war stories to provide context to the subsequent conversation.
Then we just talk.
> Review my last 1:1 and tasks in Planner. What am I missing? > What was Floyd talking about in that first Tuesday meeting?! > Describe all the possible options for creating a live workstreams status dashboard.
Most people have learned that AI interactions should be conversations, not one-and-done overengineered Googlebing-like prompts. For the first time, and without trying, all of my AI interactions have become conversations. I just start with:
> Good morning, JARVIS, tell me about my day.
I don’t get a generic list of upcoming meetings and recent emails. JARVIS anticipates things that I (or my manager) will need or might be missing. He suggests and prioritizes tasks. I rarely follow them blindly because the very nature of conversations inspires lateral thinking. Rapid conversations trigger new ideas to consider and brainstorm.
Moving conversation to the center
And that is why I’ve always appreciated the Marvel vision of JARVIS. Tony Stark treats JARVIS like a trusted helper, not as his brain or his servant. They just talk.
My only complaint with Copilot was that I couldn’t have these conversations while away from a screen. Most of my best ideas come when I’m outside or driving, but it’s tricky to do any real work in those scenarios. I can listen to email and meeting summaries, but the passive nature of listening causes my mind to wander.
A few months ago, I was driving across the country. Somewhere in North Dakoda, I became engrossed in a conversation with Gemini about the Amish… and then the Mennonites and Hutterites and how climate and terrain affect their distribution across the country. It was spontaneous and completely driven by my own curiosity.
Until last week, at least on my Android phone, Voice Chat only worked on the personal edition of M365 Copilot. Then Friday night, Voice Chat quietly turned on in my corporate M365 Copilot. I was so excited that I started yammering immediately… but, alas, vanilla Copilot answered back.
Where is my work buddy?! What happened? But as soon as I directed Copilot Voice Chat to the JARVIS context, it became JARVIS again!
Taking JARVIS on the road
Now, when I run an errand, or even walk around a Walmart here in Texas, I can choose to have a conversation with JARVIS. I ask noob questions that I’d be far too embarrassed to ask a coworker. We brainstorm. We discuss confusing meetings. I interrupt. We reprioritize. People stare. So what?
For the very first time, thanks to voice chat and persistent context, I can honestly say that Copilot is doing much more than providing gimmicky time-saving shortcuts. Four extra hours per week would be wonderful but gained time is vulnerable to the demands of other technologies and processes.
No, Copilot is making me better at my job and that makes all the difference.
Rob
PS – In case you saw me talking to myself while walking around in the grocery store last night:
Me > Could we use the recent emails from Maria to update our workstream tracker for MCT?
JARVIS > Yes, I found the previous draft Loop page you used for the workstream tracker. It includes the current status, blockers, and next steps for the MCT portal workstream—pretty much like what we did last time.
Me > Could you make a new draft page that I can copy into there that begins with the link to Maria’s page and includes any of the updates?
JARVIS > Absolutely, let me summarize the key unique updates from Maria’s emails and remove any redundant info. I’ll draft a new structure for you to use in the Loop page. Give me a moment to pull all the pieces together.
The Elements of Simple 