I'm not making recommendations yet. These are my deep read into the conversations and your January interviews: where the team lines up on AI POV, where it splits, and what you're running on today.
Quick pre-read ahead of our call.
Archived as read on Aug 3. The living version is Where we're at.
My conversations with Avery and Kevin over the past two months, the January 2026 team interviews, and the AI policy you already wrote. The evidence base is six months old in certain places, so things may have changed since then fairly dramatically, which we can pick up in our Monday chat. If you'd like to add or edit the agenda, shoot me a note at aggarwal.shabnam@gmail.com
Research is the use everyone already uses. Primers, paper summaries, landscape reads, meeting prep. Nearly everyone who uses AI values this first. It's the one workflow that everyone's aligned on as valuable.
Nearly half the team named data privacy concerns. 6 of 14 raised it: not just "is it safe" but what have portfolio organisations actually consented to? Your policy already draws the line at proprietary data in non-enterprise tools, so the concerns and the policy agree, which is great.
People asked for direction on the record. Multiple interviews said individual choice isn't working and the organisation should set expectations, which is rare, but it also fuels this work beautifully.
The builders and the cautious users want the same things. This one surprised me. The heaviest builder and the most careful non-user landed in a similar place:
"Big risk of increasing inconsistency if people aren't working in the same container."SANAT · DAILY USER
Ownership around systems is needed; it shouldn't be up to each person.KRISTIN · LIGHT USER, PARAPHRASED
Tool loyalty is personal and sticky. Five of you pay for Claude, three for ChatGPT, one lives in Gemini. The reasons behind these choices seem to be about relationship as much as about capability:
"It knows me now."SANIA, ON WHY SHE WON'T SWITCH
"Already in my workflow. Google already has my data."KENICHI, ON GEMINI
At least one of you would rather leave a vendor over their politics, and more than one distrusts a company whose tool they use daily. So a consolidation conversation here might be a values-and-effort conversation ahead of anything else - but it may be the case that we don't consolidate across one tool as well.
Writing help splits the room. Some of you draft with AI daily and want more. Others found it overedits, inserts errors, and produces text that reads as obviously machine-made. One early experience of an AI-added error in a memo cost lasting trust that hasn't recovered yet.
Free and personal accounts sit awkwardly against the privacy consensus. Several accounts in regular use are personal or free tier. The same team that identified privacy as its top concern is partly running on accounts the organisation can't see or switch off or control, which brings up a data privacy and security concern.
In their own words, all from people who use these tools every day:
"AI is making me dumber, not retaining as much."KENICHI · DAILY USER
"Writing is thinking."ALEX · DAILY USER
Worry that writing becomes "anodyne and dull."KEVIN · DAILY USER
"Authenticity and slop are hazards."CHRISTIAAN · DAILY USER
Assistants first: Claude on a team plan with two premium seats that are hitting daily caps, one admin ChatGPT account, one admin Perplexity account, Gemini inside Workspace, and enterprise Fathom for meeting notes. The policy already bars proprietary data from anything outside these.
Underneath the assistants sit the systems that hold Mulago's actual knowledge. The table below is the state of play for connecting AI tools to each one properly, through a live connection.
| SYSTEM | WHAT IT HOLDS | LIVE CONNECTION AVAILABLE? | ACCESS MODEL | FLAG |
|---|---|---|---|---|
| Box | Portfolio files, org intel, financials received | Yes, vendor-provided | Per-person sign-in; read and write possible, start read-only | Today's tools were trained on copies of Box files, and copies go stale |
| Google Workspace | Fellows-side files, mail, calendar | Yes, with limits | Per-person sign-in; more restricted than most vendors | Reported as harder to set up than expected; test early |
| Airtable | Sourcing pipeline, three-plus years of candidate records | Yes, vendor-provided | Per-person or key-based; read and write, start read-only | The richest structured data you hold; several January wishes point here |
| Salesforce | CRM, exclusion lists | Yes, vendor-provided | Per-person sign-in; scope tightly | Exclusion lists feed screening, so accuracy here matters most |
| Slack | Team communication | Yes, vendor-provided | Per-person sign-in | Lower priority; useful later for search and posting |
| Fathom | Meeting transcripts and notes | None known | Export only, most likely | Under review on your side already; fine to leave unconnected |
| Bill.com | Payments | Not applicable | Deliberately none | Not to be used with AI |
Two honest flags on this table. "Yes" means the vendor maintains an official connection, which matters because anything built in-house requires in-house maintenance. And every "start read-only" is deliberate: reading is where nearly all the early value is, and writing can wait until trust is earned.
Some questions I'd love to ponder in our kick off chat:
This should be a task that eats up your week and has a pattern to it.
This might be something you tried that didn't work as you'd hoped, or that made you trust the output less. This is as useful to us as the successes, since it's vital we trust the stuff we're going to build and rely on.
This is somewhere you'd prefer one clear organisational answer instead of deciding for yourself each time, or for your colleagues to be "using judgement" instead of leaning on a rule.
Excited to kick this off Monday!
See you Monday.
Shabnam