Ryan Rhodes
84 posts
I wanted to experience the extreme end of agentic programming, so I went to a gastown workshop. I don't know whether it's the future. I do know that I've never seen an ecosystem with docs so consistently wrong, or terminology apparently designed to confuse and infuriate.
English
I have a suspicion...
we should talk to AI using code instead of text.
even descriptions of things should just be code.
much more clear & concise.
English
@scottdwitt @davidsenra @elonmusk how many people actual clean up their own strategic mess?
harder to do this than start from scratch.
English
@davidsenra @elonmusk Wasn't Elon the founder of Starlink?
Seems like he was charging-in and cleaning up his own strategic and managerial mess (aka: FounderMode).
Rethinking the previous approach is smart, but maybe not heroic.
English
How @elonmusk fixed Starlink:
“Starlink was a mess. It was 10X too expensive and they were building 1/10 of how many they needed.
Elon’s like I've had it. This is now the bottleneck. I'm fixing this.
He grabs a team of engineers that he trusts and they fly up to Seattle.
They fire the entire Starlink leadership team.
They sit down in a war room and they start running the algorithm.
•What is the first principles of satellite design?
•How simple can we make this thing?
•Why does this exist?
•Why are these two things so far apart?
•Why do we need this much energy?
•Why do we need this manufacturing process?
And over the course of a few months they make a two order of magnitude leap.
These people had never encountered this design before, but just by applying the algorithm and working with maniacal urgency towards this extremely high design bar, they created this product that's now —if it was a standalone business —would be worth tens of billions of dollars [or more].”
David Senra@davidsenra
My conversation with @EricJorgenson, author of The Book of Elon (@elonmusk). 0:00 Book Reveal 0:39 Build Useful Things 2:19 Engineering Talent Edge 4:26 Wired for War 6:47 Tip of the Spear 8:47 Burn the Boats 13:13 Facing Fear 15:16 Origin Story Myths 18:19 Know Business A to Z 22:17 Simplify and Fail Fast 25:35 Reality and Physics 28:18 The Algorithm Begins 30:34 Delete and Simplify 34:25 Starlink War Room 36:52 Repetition as OS 38:18 Step Three Simplify Optimize 38:43 Question Every Requirement 39:13 Tesla Battery Pack Delete 40:43 Repetition Installs Ideas 42:02 Step Four Accelerate 43:26 Design Org for Speed 46:06 Step Five Automate 46:29 Control and Clean Sheet 48:54 Vertical Integration and Costs 50:47 SpaceX Incentives and Mars 57:11 Frontier Unlocks Starlink 1:00:26 Time as True Currency 1:03:58 Speed Triage and Bottlenecks 1:10:11 Internalized Responsibility 1:12:56 Avoid Serialized Dependencies 1:14:31 Aligning the Team 1:15:07 Time Is the Constraint 1:16:00 One Metric Focus 1:18:03 Directional Predictions 1:19:06 We Must Make Stuff 1:25:39 Manufacturing as Moat 1:26:23 Speed and Direct to Customer 1:28:41 SpaceX Feasibility Study 1:33:07 Edge of Sanity Leadership 1:37:10 Bottlenecks and Integration 1:40:01 Design and Simplify 1:45:15 Catch the Rocket 1:48:14 Capitalism and Closing Includes paid partnerships.
English
@andrewmccalip @dosco i'd love to hear it! what is your 10 minute explanation.
English
@dosco yeah it didn't click for me either the first time. I've been practicing the explanation to try to get people there in 10 minutes.
English
@mslipper @jeremie_strand that is an AI response you responded to
English
@jeremie_strand What perf concerns have you seen? With iron-proxy you're talking sub to single-digits of additional latency in return for dramatically better security. The workload would have to be very perf-sensitive to trade off the security benefits.
English
IMO putting secrets in env vars is an antipattern. The Vercel thing makes that clear. Workloads shouldn't have secrets at all. Keep them in a vault instead and have an egress proxy inject them on the way out. This will seem obvious in a year.
English
@sandislonjsak why areyou trying to sell people onthings they dont want?
if they dont want to use it, help them do something they want to do.
English
Today I heard people don't want to invest into self hosted models on prem because it will be obsolete anyways in 2 years.
I have no idea what to think about this line of reasoning...
What do you guys think? Does this make sense to you?
English
@DonnyWals Ive been diving deeper to understand the pieces of the harness engineering thing.
I think each workflow will be so specific to each job.
the coding agents arejust the beginning.
English
To everybody that was absolutely on fire with their agents back at the start of the year: how are you doing right now?
Personally, I've found that the novelty has worn off and agents are just another tool in my toolbox instead of making me want to run them 24/7 like I wanted in January
English
@hiiinternet honestly how many people are there that fill these requirements?
English
Hiring for over 100 startups backed by investors like a16z, sequoia, yc, gc, etc.
One of the roles in highest demand is AI engineer.
The engineers who do best on Fonzi:
- Have built AI-powered products end-to-end, not just prototypes, but systems handling real users
- Can talk fluently about retrieval, memory, agent frameworks, context engineering, eval pipelines and anything else at the cutting edge
- Combine strong software engineering fundamentals with deep awareness of what's happening at the frontier of LLMs and agentic systems
- Thrive in startup environments where the problem is ambiguous and the team is small
- Have 2-8 years of experience, ideally with experience at at least one high-growth company
if this is you you should onboard, get bids from startups and talk to the ones you're interested in.
x.com/i/jobs/2040114…
English
@7N8Nine @taobanker the AI psychosis will only get worse though.
and those people won't go back to "learning the basics"
theyre on crack they arent gonna eat healthy now.
English
@taobanker I almost view it as the opposite of a bubble. People's AI psychosis is preventing them from properly utilizing it. New tool and people are just barely starting to get how to use it without losing their mind
English
Everyone who is skilled and experienced at using these tools comes to the same conclusion.
IMO this fact alone is the strongest evidence that we are in a bubble -- people have irrational beliefs that will mellow out, they are just 6 months behind the curve of adopters.
fj@fjzeit
my entire career strategy hangs on a strong belief that we are not going to see fully automated software generation in my lifetime. but we are going to see an end to the ralph-loop, spec-driven-one-shot-dream, and the "end of white collar work" hype. we're already seeing some high profile players in software development start to set their coarse along the same path i've been following these past few years. it starts with "wow" then "i can use my skills to fully automate this" and then it proceeds to "fast but no cognitive ownership" then disappointment, confusion, frustration, and ends with "hey, this isn't going to work guys, we need to be more disciplined and look at the code, keep our cognitive ownership, and just use the tools to improve our outcomes. these are not our replacement, these are our accelerators. it's the same story, different tooling". i'm already there. if a company were to come to me today and say "we tried all the trendy stuff but it just made everything worse, we're losing control of our code base, we need to either ditch these things or make them a power-up" then i am ready for that. if i am wrong, then so be it. my career is over anyway because i have zero interest in giving up cognitive ownership and responsibility (the ability to respond), while remaining accountable. i'd rather wash dishes or stack shelves than submit myself to the horrors of remaining accountable without cognitive ownership, agency, and responsibility.
English
i'm thinking about this because no matter how much you try, the existence of LLMs just makes our codebases rot more
i feel like having a few people just cleaning up every day without getting pulled into feature work could be good
assuming they enjoy it and don't get burnt out
English
@neural_avb so it can read arbitrarily long texts using code.
run subagents to perform tasks that respond back with adding variables to main repl. saving context
and answer the user with a FINAL(var_name) so even if var_name is long it wouldnt pollute the context?
super context management
English
Yeah its a bit tricky to process
Can you read this article? It goes over the difference between CodeAct, ReAct and RLM with a very simple example. I am 100% sure it will clear your main doubts.
Feel free to ask questions after that, I'll be happy to answer
x.com/neural_avb/sta…
English
What is the most legitimately exciting thing you've seen in the engineering AI space that hasn't yet gone mainstream?
English
AI is teaching us that context digestion is more important than IQ.
An IQ test tests how fast and accurately humans can respond to very short prompts
But impressive human or AI performance in the real world has more to do with how we can synthesize vast quantities of context.
We need a CQ test
English
Oh damn you guys REALLY liked this video
What other AI topics and terms should I try to demystify next?
Theo - t3.gg@theo
Agent harnesses aren't the black magic many of y'all seem to think they are. To prove it, I built one.
English