Logbot (Mar 28 2025 at 19:01): Logbot (Mar 28 2025 at 19:01):Long-time JS Party panelist Amal Hussein joins Jerod to catch up on her career path, to opine on the viability agentic coding, to feel all the feelings that AI brings out of us as developers, and to share something new in her life that changes everything. :link: https://changelog.am/86
| Ch | Start | Title | Runs |
|---|---|---|---|
| 01 | 00:00 | Let's talk! | 00:38 |
| 02 | 00:38 | Sponsor: Heroku | 02:43 |
| 03 | 03:21 | Big features & Friends | 02:37 |
| 04 | 05:58 | Did something go wrong? | 04:08 |
| 05 | 10:06 | Amal's journey | 04:09 |
| 06 | 14:15 | Her time at Stripe | 02:37 |
| 07 | 16:52 | Her time at Cisco | 02:29 |
| 08 | 19:21 | Amal the disruptor | 02:37 |
| 09 | 21:58 | Silent sacking | 03:03 |
| 10 | 25:01 | Sponsor: Depot | 02:20 |
| 11 | 27:22 | Where she works now | 08:02 |
| 12 | 35:24 | How she found Istari | 03:23 |
| 13 | 38:47 | Moving to a different platform | 02:10 |
| 14 | 40:56 | Istari agents | 03:38 |
| 15 | 44:35 | Every deployment target | 01:31 |
| 16 | 46:05 | Do agents work today? | 03:49 |
| 17 | 49:55 | Agentic coding | 01:19 |
| 18 | 51:13 | Sponsor: Retool | 03:01 |
| 19 | 54:14 | Amal's AI feelings journey | 02:39 |
| 20 | 56:53 | Jerod's agentic coding opinion | 04:09 |
| 21 | 1:01:02 | Humans out of the loop? | 01:20 |
| 22 | 1:02:22 | Jerod is basic | 01:17 |
| 23 | 1:03:39 | Software is hard tho | 01:58 |
| 24 | 1:05:37 | Cheering the demise of engineers | 04:04 |
| 25 | 1:09:41 | We've been a bottleneck | 04:32 |
| 26 | 1:14:13 | Vibe coding gold rush | 00:56 |
| 27 | 1:15:09 | Revenge of the juniors | 04:46 |
| 28 | 1:19:54 | A basic analogy | 00:59 |
| 29 | 1:20:53 | It's just a job | 01:28 |
| 30 | 1:22:21 | Amal had a baby! | 05:52 |
| 31 | 1:28:14 | Scaling out | 01:31 |
| 32 | 1:29:45 | Don't wait | 03:22 |
| 33 | 1:33:07 | Why wait? | 03:18 |
| 34 | 1:36:25 | Bye, friends | 00:25 |
| 35 | 1:36:51 | Closing thoughts | 01:01 |
Don MacKinnon (Mar 31 2025 at 13:58):Congrats on the baby Amal! Was wondering why she wasn't around as much on JS Party over the last year, now it makes sense.
Matthew Sanabria (Mar 31 2025 at 17:40):Love the kids chat at the end. We need more of that.
Jerod Santo (Mar 31 2025 at 17:41):Hah we saved it to the end because we thought folks might want less of that...
Matthew Sanabria (Mar 31 2025 at 19:20):To be fair most people probably do want less of that because they just want to hear tech all the time but it's nice to hear the human side of things. Especially when people say they would have had kids sooner.
valon-loshaj (Apr 01 2025 at 11:28):loved the conversation, tech and non-tech, congrats on the new bebe and on what sounds like an amazing role amal!
Jimmy Sweeney (Apr 05 2025 at 20:33):Regarding what we can expect from agentic coding, I think Amal’s point about a limit to human bandwidth is exactly right. Something I’ve thought about is how my team and I are already nearly at a bottleneck given how much code we have to review. If 10 more code reviews came in each day because we are all just chucking out vide code, who’s gonna review all that? I suppose there might be a future where we are okay with AI agents reviewing AI generated code… sort of like what Jerod was saying about LLMs checking LLMs for hallucinations… and that is an interesting future to consider.
One thing I’d be all for an AI agent helping me out with, though, is doing my debugging for me. I’d love to be able say “here’s the problem I’m having, go take the whole code base as your context, go interact with the app in the browser, do whatever you need to do, and give me a summary of what the bug is while I have a cup of coffee.”
Tim Uckun (Apr 05 2025 at 22:01):It may not be an AI but I do think there will be more/better review tools. I am thinking of something akin to a linter or a type checker but better and more context/code aware.
I am also on the record for predicting we will be using a AI specific programming language in the future. Something along the lines of rocq (coq) which is a language that is provable. The AI will write the code and then run the compiler which will error out if it's not provable. The language I envision will also include things like fuzz testing and such so that when the AI writes the code it will also write a full test suite which passes.
If this language is sufficiently concise and easily readable like ruby then all you have to do is check the business logic and you can be assured the rest of the code is solid.
The next step after that will be to create a language incomprehensible to humans but safe and easy for the AI :grinning:
Ron Waldon-Howe (Apr 06 2025 at 00:38):LLMs are also only a small part of the AI field
We'll eventually hit a local maxima of what works well with generative AI (given that it has no knowledge and only accidentally produces useful output, provided an expert actually reviews the output)
And when that happens, we'll probably see universities and companies return to the broader field and develop something better, hopefully something that internalises facts and thus doesn't fabricate
Tim Uckun (Apr 06 2025 at 01:09):LLMs were trained on a broad set of things including poetry, fiction etc. I could imagine some that are only trained on a particular language and nothing else. They would receive structured input (not free text) and then spit out only that language and would be extremely competent.
Last updated: Jun 28 2025 at 15:16 UTC