Adam Jacob joins us to discuss how agentic systems for building and managing infrastructure have fundamentally altered how he thinks about everything, including the last six years of his life. Along the way, he opines on the recent AWS outage, debates whether we're in an AI-induced bubble, quells any concerns of AGI and a robot uprising, eats some humble pie, and more. :link: https://changelog.fm/664
| Ch | Start | Title | Runs |
|---|---|---|---|
| 01 | 00:00 | This week on The Changelog | 01:13 |
| 02 | 01:13 | Sponsor: Namespace | 01:42 |
| 03 | 02:55 | Start the show! | 01:01 |
| 04 | 03:56 | The AWS outage | 03:41 |
| 05 | 07:37 | The public's response | 03:24 |
| 06 | 11:00 | The cloud exodus | 04:31 |
| 07 | 15:32 | Taking control back | 05:18 |
| 08 | 20:50 | Who will build it | 05:43 |
| 09 | 26:33 | Maybe not a bubble | 05:15 |
| 10 | 31:49 | Sponsor: Tiger Data | 01:38 |
| 11 | 33:27 | Robot uprising concerns | 02:10 |
| 12 | 35:37 | Back to the practical | 03:51 |
| 13 | 39:27 | AGI locked in the basement | 01:48 |
| 14 | 41:15 | Real-world value | 01:44 |
| 15 | 42:59 | From the horse to the car | 02:57 |
| 16 | 45:56 | Changing Adam's mind | 03:44 |
| 17 | 49:40 | Humility required | 02:48 |
| 18 | 52:29 | Infra people resist | 03:50 |
| 19 | 56:19 | Practical AI slop | 06:23 |
| 20 | 1:02:41 | Sponsor: Notion | 01:48 |
| 21 | 1:04:29 | Babysitting or nah? | 03:13 |
| 22 | 1:07:43 | Improving the loop | 01:51 |
| 23 | 1:09:34 | A post-AI language | 03:59 |
| 24 | 1:13:33 | Paying the piper | 01:55 |
| 25 | 1:15:29 | The agent is the glue | 03:23 |
| 26 | 1:18:52 | System Init's sauce | 03:35 |
| 27 | 1:22:27 | What 'custom model' means | 02:18 |
| 28 | 1:24:44 | Humans in the loop | 03:21 |
| 29 | 1:28:05 | The MS-DOS era | 03:49 |
| 30 | 1:31:54 | Positioning System Init | 03:51 |
| 31 | 1:35:45 | The cloud OS? | 03:55 |
| 32 | 1:39:40 | The on-prem situation | 03:30 |
| 33 | 1:43:10 | Selling more than hiring | 09:22 |
| 34 | 1:52:31 | The sales approach | 02:52 |
| 35 | 1:55:24 | Every single deal | 06:30 |
| 36 | 2:01:54 | Wrapping up | 00:27 |
| 37 | 2:02:20 | Closing thoughts | 01:18 |
This episode was fantastic! I've already started listening to it again. I've though that SI was a good way for people who aren't experts in infra to build and learn infra, and I think this method is great. Now, I just need to actually use it
A language made for AI programming would probably have to be some combination of Haskell and Roq. It would most likely be procedural or functional and not object oriented. It would most likely be relatively hostile to humans and overly verbose. It would have only one way to do anything. It would have to compile fast which means pascal like forward declarations of everything i.e a var section at top of the function where you declare all the variables you are going to use and any function you use must have at least the interface of it declared before you use it.
That's all presuming it has to be a human readable language as opposed to something like a VM or assembly language because theoretically the LLM would just output assembly or even bytecode.
It would have a very rich type system something like ADA where you can declare subsets of numbers i.e type Grade is range 0 .. 100; and even units like 5kg where the compiler can catch violations like assigning 500 to a grade or adding kilometers to kilograms.
I also think the language would come with a spec language for prompting similar to cucumber and a testing language which is super high level and human friendly so humans can write specs and tests to make sure the LLM is doing what it should.
@Tim Uckun I get nice Haskell vibes from https://www.unison-lang.org/
Unison has the same business model as Deno: they offer cloud hosting as "the best way" to run their language/runtime
@Trevor Settles happy to hear you enjoyed it! What, in particular, makes this one worth a second listen in your opinion?
Last updated: Dec 16 2025 at 01:26 UTC