On this page
Make your glyph
Type a name and the four shapes behind every build's mark rearrange into a glyph of your own, ready to download as an SVG. No AI and no cost: the same name always gives the same glyph.
The lab of Emily Sas
What do you learn about AI by building with it?
I'm a product leader who builds with AI agents, and this is my playground for finding out: what I built and how, what it cost, what broke, and how sure I am of each figure.
Working on nowMullo
Try it
Everything you can tryInvite-only beta
Fold
A recipe app for vegetarian households with a vegan at the table. Every recipe shows how it could be made vegan where possible, where to split the pan and how much each of you needs, so one meal works for everyone. If you cook for mixed diets, ask for an invite and you'll hear from me when the beta opens.
Build logs the story behind each one
All builds- EXP-009Building
Mullo
The best page, the worse journey: when an AI tool's quality isn't worth the effort
An AI that draws my notes as an illustrated page. Against a general-purpose tool it made the better page but took far more effort, so I'm redesigning it around what only it does.
- EXP-007Building · Pilot
This lab, rebuilt
Nothing goes live unread: building a public site with AI agents, without losing control
This lab, rebuilt in three days by Claude Code from briefs I worked out with Claude. The setup that keeps an agent-built site honest, and the rules that came from mistakes.
- EXP-006Building · Pilot
Fold's review panel
Never mark your own homework: independent AI reviewers for a solo builder
Independent AI reviewers for an app I was building faster than I could read its code. They caught bugs that had passed the tests, and one finding changed what Fold is.
- EXP-005Building · Pilot
Fold
Cheap to build, hard to stop: when AI builds faster than you can learn
A recipe app for vegetarian households with a vegan at the table, built for my own family's needs; once I saw its wider potential, I added a fair amount of complexity without checking whether it was warranted.
- EXP-004Live
emilysas.com
The bottleneck moved: where the time goes when an AI agent writes the code
A Claude agent wrote my professional site in about 42 hours, from first commit to the last launch change. The time went on accounts, decisions only I could make, and testing in production.
- EXP-002Retired
Launchling
All tests passing: when checking an AI product outgrows the product
An AI tool that turned a founder's rough idea into a plan. The product ran end to end; the evaluation I built to check it grew until it couldn't finish a run.
Questions I'm working on
All threads- Breaking it on purposeHow do you find the ways an AI product fails before your users do?2 pieces
- Knowing if it's any goodWhat does “good” mean for output nobody can mark right or wrong?6 pieces
- Checking the checkersWhen AI checks AI, what checks the checker?5 pieces
- Building with agentsWhere do a builder's time and judgement go when agents do the typing?7 pieces
- What it knows and can reachWhat should an AI know and be able to reach, and who decides?2 pieces
- Seeing the costWhen is a cheaper model the better product decision?7 pieces
- Designing for trustHow do you design AI that people trust exactly as much as they should?7 pieces
- Is it worth buildingWhen building is this cheap, how do you decide what's worth building?10 pieces
Latest writing
All notesNote · 3 threads
Diverse by prompt, biased by role: how AI illustration passes society's stereotypes on
I asked for more diverse people in the sketch notes an AI pipeline draws for me. The AI writing the scene descriptions obliged by attaching an identity to each scene, and in doing so gave a wheelchair user a stage to conduct from, and the cleaning to an older Black man, alone, scrubbing a heap. The image model drew exactly what it was told, so every check passed.
Guide · v2 · updated October 2026
How I evaluate AI output
A living guide. I wrote v1 in 2025 after building Launchling's evaluation framework; v2 is how I'd do it in October 2026. I've kept both, because the gap between them says as much as either version does.
Note · 4 threads
A recipe for honest pictures: how to avoid AI image slop without copyright trouble
People choose what to cook by picturing it, but cookbook photos aren't mine to reuse, and AI pictures of dishes nobody has cooked are usually wrong. Here's how separating looking from drawing gave every recipe in Fold a picture that's true to the dish without copying anyone's.
Keeping pace
Full timelineAI in the world
What I built
What I learned
Building with AI since 2022, set against what changed in AI itself. The full timeline has the detail behind every point.