A year and a half ago, Hannah Hoffmaster was using AI to look up high-protein snacks. Today she’s running multi-agent systems she built herself, with no prior coding background. Bridging the gap between those two points was the Foster School of Business’s Masters of Science in Information Systems, or MSIS program, and few specific classes.
I spoke with Hannah about her experience for my podcast series, AI Ready, in the spring of 2026.
Rating yourself honestly is the starting point
I asked Hannah to rate her pre-Foster comfort with code and AI on a scale of one to seven. She landed on two. She’d dabbled in R for a statistics class and didn’t know what GitHub was. Then she took Professor Leoonard (Leo) Boussioux’s course at Foster’s MSIS program and built a working website in an hour, for homework, after a previous solo attempt had taken her a month.
That gap — a month of struggling alone versus an hour with the right tools and guidance — is the entire pitch for why non-technical professionals should stop waiting until they “know how to code” to start building.
The lesson that took a few repetitions to land
Hannah’s first build was GiveWise, a charity-navigator website with scored profiles. When a showcase attendee asked how new charities get added to the platform, she realized her answer — “they’d email me” — wasn’t actually using AI at all. So she built a three-agent pipeline: one agent scrapes identifying data on a new charity, a second formats and scores it consistently with existing profiles, a third reviews the output and either flags anomalies to Hannah or emails the charity directly for missing information.
It sounds clean in retrospect. Getting there wasn’t. “One of the issues was that I was just trusting AI. I’d tell it what I wanted, it would say ‘I did that,’ and I’d move on. Later I’d come back and be like, what’s this hot mess?”
The specific failure: every charity profile showed the same low outreach score. The model had quietly benchmarked the people each charity serves against the entire global population instead of against the actual target population of the charity — a reasonable-sounding shortcut that produced meaningless numbers across the board. She only caught it because she went back and checked, profile by profile, instead of assuming the output was correct because it looked finished.
Her rule now: “You’ve got to be the one thinking about the workflow. You’ve got to be the one thinking about the processes, and then having the AI do the work for you that you don’t want to do.” That’s directed delegation — you still own the logic of what “correct” looks like.
Building the tool you actually need
Hannah also built Offload, a scheduling app for people with ADHD, after recognizing that most planning apps fall apart the moment one block of the day shifts. The chatbot inside it, Nudge, renegotiates your schedule in real time — shortening task blocks when your energy crashes, reordering for harder work when you’re spiking, coaching you through a condensed routine when you’ve lost track of time and need to get out the door.
The pattern worth noticing: she didn’t set out to build a product. She built a fix for her own friction, then noticed other people had the same friction. That’s a more reliable startup engine than starting from “what’s a good AI idea.”
What community adds that solo experimentation doesn’t
Hannah builds for fun — she’s in Claude Code on weekends, not because an assignment requires it. But she’s explicit that the bigger unlock came from other people: “When you get the other perspectives of your peers, of your classmates, of these other people who are excited along with you, then you get different perspectives. The possibilities with AI are borderline endless.” One person prompts a certain way because it works better for their use case; you’d never stumble onto that alone.
Her advice to incoming students doubles as advice to anyone starting from zero:
- Don’t wait for technical fluency before you start building. The tools compress the learning curve more than you expect.
- Verify everything, especially outputs that look complete. “It did that” is not the same as it actually did that correctly.
- Build the thing that solves your own problem first. The market validation often follows.
- Show up to the community — office hours, AI clubs, peer builders. The shortcuts you need are sitting in other people’s heads.
The takeaway: the technical barrier to building real software has collapsed. The judgment barrier hasn’t. If you’re starting from scratch, that’s good news — judgment is something you can train regardless of your coding background, starting on day one.