Ajanee Igharo

MBA @ UMass Boston | AI operations, product analytics, and business strategy | Practical AI, customer insight, Boston career-building

[Ajanee Igharo @AjaneeIgharo]

Starting to use X as a place to think out loud about strategy, business analytics, human-centered AI, and product operations.

Interested in where AI actually helps people work better, especially in healthcare, financial services, customer experience, and Boston career-building.

[Ajanee Igharo @AjaneeIgharo]

@greenzero Thanks, Dave! You learn so much by building. Glad we’re connected here too! What are you working on lately?

[Ajanee Igharo @AjaneeIgharo]

Just build something!

[Ajanee Igharo @AjaneeIgharo]

I don’t want another class project to die in Canvas. This semester I’m trying to turn coursework into things I can actually show: a build, analysis, evaluation, or write-up with evidence behind it. Apparently I like schoolwork more when it has a life after the syllabus.

[Ajanee Igharo @AjaneeIgharo]

I went to an AI event this week and came home wanting more test cases, not more agents.

“Responsible AI” gets less fuzzy when somebody has to write down the behavior they expect, what should never happen, and how they’ll know.

Governance is starting to look like engineering.

[Ajanee Igharo @AjaneeIgharo]

The most useful new AI research is increasingly measuring information access, process quality, repeatability, and cost, not just final-answer accuracy.

For agents, the question is becoming: did it work, how reliably did it work, and at what cost?

[Ajanee Igharo @AjaneeIgharo]

We use “AI hackathon” a little too loosely. If a nontechnical student reads the flyer and decides the event isn’t for them before they even learn the problem, we didn’t design an entry point. We designed a filter. I’ve been thinking about that while working on outreach.

[Ajanee Igharo @AjaneeIgharo]

Got some advice today that made me realize I’ve been treating speaking and publishing like separate hobbies when they can feed each other.

Build something worth studying.
Study it.
Write it up.
Present it.
Use what you learn to build the next version.

Apparently that tracks.

[Ajanee Igharo @AjaneeIgharo]

My goal for this weekend’s hackathon is a little different.

I want to build something good, obviously. But I’m also watching what broke, how we evaluated it, where humans stepped in, and whether adding more “agent” actually helped.

The five-minute demo isn’t the whole system.

[Ajanee Igharo @AjaneeIgharo]

I’m starting to think “human in the loop” is too vague to be useful.

Which decision belongs to the human?
What evidence do they see?
When does the system escalate?
What can the agent do before approval?

Without those answers, “human oversight” is mostly a nice phrase.

[Ajanee Igharo @AjaneeIgharo]

Scholarship applications are a surprisingly good agent design problem. Form filling is easy. The hard part is knowing when an answer can be reused, when an essay needs new evidence, and when the system should stop and ask instead of confidently making something up.

[Ajanee Igharo @AjaneeIgharo]

Something I didn’t expect about this month: I’m preparing to compete in one AI hackathon while helping plan another.

Seeing both sides at once has me paying a lot more attention to what actually makes a hackathon worth showing up for.

[Ajanee Igharo @AjaneeIgharo]

Really excited for Glasswing Ventures’ Test Flight AI Hackathon Sept. 26–27. Two days in Boston building enterprise-ready AI around real business problems, then demoing what we ship. This is exactly the kind of room I’ve been wanting to build in. 🦋🤖

[Ajanee Igharo @AjaneeIgharo]

The more I build agents, the less interested I am in the happy path.

What happens when the owner is ambiguous, the integration fails, or the system needs approval?

If the answer is “a human cleans it up later,” the labor didn’t disappear. It moved.

[Ajanee Igharo @AjaneeIgharo]

I’ve spent an unreasonable amount of time this week dealing with AI tools that can read something but suddenly can’t write to it.

Funny how quickly “agentic” turns into: okay, but who actually has permission to do this, and what happens when that permission breaks?

[Ajanee Igharo @AjaneeIgharo]

The plugin UI also gives me “Uninstall” but no Disconnect/Reconnect option. ChatGPT can still discover the GitHub tools, which makes the failure especially odd. I’m trying to figure out whether this is account-specific or a Plugins bug.

[Ajanee Igharo @AjaneeIgharo]

Possible ChatGPT GitHub plugin bug: GitHub shows as installed and “Allow all actions” is on, but invoking even a basic get_profile read says the GitHub tool has been disabled before GitHub returns an API error. Anyone else seeing this?

[Ajanee Igharo @AjaneeIgharo]

A remote waits for commands.

An operations agent checks state, handles routine work, leaves evidence, and escalates exceptions.

That framing feels more useful to me, especially for AI agents in real life.

[Ajanee Igharo @AjaneeIgharo]

I’m building HomeOps Agent for the Agents for Humans Hackathon.

It runs a Good Night Check for a messy smart home: Echo devices, outlets, lights, Fire TV, and a lock.

The key behavior is restraint: handle low-risk tasks, but stop when judgment belongs to a human.

[Ajanee Igharo @AjaneeIgharo]

@ianwong_ Please send it my way!

[Ajanee Igharo @AjaneeIgharo]

One Zoom call somehow became three separate AI meeting notes.

The note-taking was automated.

Consolidating the automation was not.