One shared home for what we build together.
Deliverables show up here as they come. Starting this week: read where we're at for what I found, then the layer underneath for what I'd suggest we do about it. Everything else is reference, so please dip in as you need it.
A live calendar link with my real openings, checked across my calendars. Pick a slot and it books straight in. The full timeline, session by session, lives on the scheduling page.
Several of you are already building real tools. Cost-effectiveness models, comparison dashboards, screening tools. In the January interviews, 7 of 14 used AI daily, 3 were ramping up, whereas 4 barely touched it. My take: let's design for that spread, ie. the middle and the edges rather than the heaviest users.
Each of those tools lives in one person's own personal setup. Which means, nobody else can find it, check it, or build on it. Two people end up rebuilding the overlapping things without realizing.
The first three months should ideally close that gap. We'll build a shared foundation, set the strategy level together, and get to two or three concrete wins on real workflows.
The first phase is internal Mulago. The fellows programme will be designed during this period with a goal to launch later - that sits on the parked list further down.
A deep look at our position: the January baseline, what the Feb guidance settles, where practice hasn't caught up, and the current stack. Updated after the system walkthroughs with Claire, Stephanie and Rahila. The Aug 3 pre-read is archived inside.
What we'll cover in the meeting, and the three things we should hopefully leave with.
Here's what we decided, what surfaced, and the actions we each took away.
Every session from kickoff to the October review, plus the booking calendar to grab time with me directly.
This is the "what should this level look like" piece we aligned on, in draft. Seven good builds, and the one foundation none of them has. Five written decisions, one folder structure, four shared pipelines and one hosting answer. Prepared for us to argue with rather than to sign off.
Every data object against every system, from today through to three possible target states: keep the split, consolidate onto Salesforce, or consolidate onto Airtable. Plus where the AI actually sits in what the team has built, which turns out to be a smaller place than it looks.
A pass through the Feb guidance together with what we've learned since: where AI writing helps, where judgment should stay human, and the retention questions still marked tbd.
A step-by-step read of the grants workflow, built from the team's own guides: which steps follow a clear logic frame, which need human judgment, and which depend on outside parties. This should ideally inform the resourcing decision as well as the architecture note.
One visible workflow built together, landing on the shared foundation rather than standing alone. Funder engagement is the leading candidate, being narrowed this week.
Practical guardrails on a single agent, built as repeatable for others. Accuracy and bias were among the most-raised concerns in January, so we'll start there.
What we got done, what we learned, and what the next phase should focus on.
This is the centre of what we agreed on Aug 3. Two places, both still to be set up. Once they exist the links will get placed here, and this becomes the first stop for teammates searching for skills, mcps and prompts.
One agreed way for any AI tool to reach Box, Airtable and Salesforce, so teams work from the same starting point. I'll design the set-up and the rules with you, and your own team can stand it up and own it with my guidance. Heavier set-up work, if you want it, can be scoped in.
How you'll use it. You'll ask your AI tool a question about the data behind a set of portfolio organisations and it reads the live records from the relevant MCP or connectors, as well as searching for existing skills to lean on, rather than spinning its wheels and wasting precious tokens.
One place the prompts, small tools and finished builds will sit, with a record of who changed what and when. This is like a shared filing cabinet with a memory. It'll have the same working split: I specify it and guide, your team runs it, so the capability remains in-house.
How you'll use it. Honestly, ideally you won't need to open it. What it holds will show up inside the AI tool you already use. Only the people building things work in it directly.
The technical names for these two are MCP connections and a shared git repository, in case that's useful to anyone who wants to read further. Neither needs to be understood deeply for it to be useful.
The show-and-tell, the grants automation map, and a first draft of the architecture note.
Agree the few workflows worth the effort, then build the first one.
Finish the first builds, put practical checks on them, design the fellows phase, and review.
Strategy, prioritization, the hands-on work with named leads, and the checks on what gets built. aggarwal.shabnam@gmail.com
Avery sponsors and is point of contact.
The fellows programme, designed during these three months and launched later. Advising portfolio organisations on where AI fits their model. Rebuilding sourcing end to end. The path from Mulago thinking to investment documents to the website. Any grants-management build: we're mapping that workflow now (see the shelf), and building for it stays parked until the map and the architecture say where a build makes sense.
These are the interesting ones, which is where this is heading. I'll keep the list here so nothing gets lost, and we can come back to it at the October review.
Enter the leadership password when prompted. (It's different from the site password for security.)