emilybache.com
Software Developer, Technical Coach, YouTuber. She/her.
emilybache.com
94 posts
2,205 followers
377 following
Prolific Poster
Conversation Starter
comment in response to
post
Anthropic's Hannah Moran finally addressed the elephant in the room at this conference when she subtly dropped "Agents are models using tools in a loop" during the intro to the "Prompting for Agents" workshop https://simonwillison.net/2025/May/22/tools-in-a-loop/
comment in response to
post
Bonus thought: switching tools as a company is really hard because you can't easily exchange boards or exports of parts of the board between whiteboarding tools.
So the cost to switch later is unfortunately quite high. Exporting and importing between different tools would be a massive feature!
comment in response to
post
It's not an acceptance test It's an approval test. You have misunderstood the purpose of this test.
comment in response to
post
The DORA research found that there is NO TRADE-OFF BETWEEN SPEED & QUALITY
Teams with good scores on both, spend 44% more time on new features!!!
Read “Accelerate” by Nicole Forsgren et al.
4/4
comment in response to
post
Escape the pitfall! Watch the video
youtu.be/tu8qaKa0pKQ
2/2
comment in response to
post
ApproveJ is a new implementation of Approval Testing for the JVM. It provides a fluent API and is highly customizable.
Check it out
👉 approvej.org
comment in response to
post
I’ve also found they are often really good at writing tests if you work in thin TDD slices, and guide it by expressing the intent of the test yourself first.
comment in response to
post
As you'd expect from Josh, it was a very measured, and I think spot on analysis of where things currently stand and where they are headed.
You can give it a read yourself here: www.joshwcomeau.com/blog/the-pos...
comment in response to
post
Auto dubbing sounds like a terrible idea for these kinds of videos where so much is about telling my story in my authentic voice. I wasn't aware of this feature and will work with the MSE team to turn it off!
comment in response to
post
The way I solve this now is by making cheating essentially structurally impossible. You can’t cheat at a “test” where the entire point is assessing:
1. Do you have intellectual humility
2. Do you work well with others
3. Do you evaluate info coherently + critically
4. Can you adapt on the fly
comment in response to
post
I have no insight into that of course. I did read an article suggesting we could have AI models for things like improving industrial processes and balancing intermittent electricity supply. Those wouldn't be done by LLMs, they'd be specialised models.
comment in response to
post
Not all datacenters are built to power LLMs. Other kinds of AI like alphafold for protein structure prediction also need them. Perhaps you are being overly pessimistic