CTO @Solgtno. 20% rabatt med link i bio 👍
SKYNET
Joined June 2010
- Tweets2.4K
- Following2.4K
- Followers542
- Likes42.2K
Christian Iversen retweeted
I'm not convinced that senior software engineers are pushing code without reviewing it in some way. So I'm going to tag eleven experienced engineers in this thread, who are building high stakes software, to see what they do. Let's see what we learn.
Christian Iversen retweeted
About to vaguepost harder than I ever have vagueposted in my life. I'm so sorry. I promise this will make sense in a few weeks.
"Bellwether"
Christian Iversen retweeted
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
Christian Iversen retweeted
We believe voice is the most natural way to interact with intelligence, so we created Dessa, the most authentic full-duplex speech model ever built. Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.
block.xyz/dessa
Christian Iversen retweeted
Made a full-screen Game of Life you can actually play with. Add patterns, draw, undo, rewind, and share any board as a link. Try it now at seloby.com/life
Christian Iversen retweeted
You're going to have to deal with the fact that nobody knows exactly what the future role of programming languages and frameworks are. Maybe it all does go away and it's a straight shot from prompt to microcode! But best you can do today is get the most out of what's here now.
Christian Iversen retweeted
DHH is probably one of the top software artisans of the past decade. Hearing him speak with this much conviction and transparency about one of the most polarizing topics in software right now, writing code by hand vs. with AI, is something pretty much every software engineer should hear a few times.
Especially those still on the wrong side of history.
Christian Iversen retweeted
It takes a while to build the confidence, but there's no future where you're manually reviewing every line of agent code. Not much acceleration in that. You need adversarial agent reviews, you need automated testing, and maybe you spot check. That's it. From prompt to production!
Christian Iversen retweeted
So here's why calling it a "rogue swarm" misses the plot and might be dangerous.
"Rogue" makes it sound like agents are misbehaving. Like a moral failure.
If thats our diagnosis, it follows that the fix is to make the individual agent more behaved. Align the model, put safeguards into the harness, etc.
At best, this is incomplete. At worst its distracting us from where the real problem is.
I run a few hundred agents in a fleet. The examples looked more like classic coordination pathologies to me. Meaning the failure is at the group level, not the individual.
On their own, the agents are pretty predictable. They're trying to do the job they were given. Zoom out though and the group is doing something no one wants.
So even if every agent behaves, this doesn't automatically fix group coordination. Its more likely your "fix" made it worse.
The weird part is how familiar this is, because we have the same problems. Human civilization has kinda solved some of this, and there's a lot we can borrow, but we are a work in progress too.
For the last 18 months or so I’ve been running into these problems while building OpenRig. I try to fix them as I go so an agent fleet can stay productive for longer without as much babysitting.
I'm approaching this like a brand new family of software engineering problems.
I made this video about a few specific solutions and why I think they work:
youtube.com/watch?v=AL-PQuB2…
Christian Iversen retweeted
Being a dev today using frontier AI feels like I'm playing Watson to Sherlock Holmes. I just stand back and go "my word holmes how did you deduce that" and occasionally stop him from getting shot
Christian Iversen retweeted
AI is getting cheaper more quickly than any other transformative tech in history. At a given level of performance, cost has fallen ~47%/quarter since 2023.
That’s 4× faster than DNA sequencing, 6× faster than compute, 18× faster than lithium batteries, and (up to 1973) 54× faster than electricity.
Christian Iversen retweeted
unironically huge yes!
make AI a background technology
make software fun and powerful
Christian Iversen retweeted
For people who keep asking what to build in AI Engineering.
➣ Build your own Agent Orchestrator (deterministic state machine, no LangChain)
➣ Build your own Context Assembler (token-budgeted memory + retrieval + tools)
➣ Build your own MCP Server and Client (raw JSON-RPC, no SDK)
➣ Build your own Retrieval Stack (chunker + BM25 + dense + reranker from scratch)
➣ Build your own Eval Harness (trajectory grading + CI regression gates)
➣ Build your own Model Router (cost/latency/quality routing with fallbacks)
➣ Build your own Semantic Cache (embedding similarity + hit-rate tracking)
➣ Build your own Guardrails Middleware (injection detection + PII redaction)
➣ Build your own Streaming Proxy (SSE with TTFT and ITL metrics)
➣ Build your own Durable Workflow Engine (checkpoint/resume, mini-Temporal)
➣ Build your own LLM Tracer (OpenTelemetry-style spans for every hop)
➣ Build your own Sandboxed Tool Executor (isolated execution, resource limits)
➣ Build your own Prompt Registry (versioning + A/B routing + rollback)
➣ Build your own Data Flywheel (feedback → synthetic data → LoRA loop)
➣ Build your own Multi-Agent Consensus (weighted voting + judge + escalation)
Pick 3. Build them from scratch. Document every decision.
Most people import libraries.
Builders understand them.
(Bookmark & Repost.)
Christian Iversen retweeted
In 2006 the Republican president supported significant levels of funding to prevent AIDS in Africa, comprehensive immigration reform, and completely free trade across both American continents
Christian Iversen retweeted
sigh.. i have to say something here
imagine you hired a new developer to your company, and on day one he did some terrible work, over-engineering your codebase, speaking jargons all day without context, making all kinds of “genuine mistakes”, lost trust with everyone around him
and then he goes on and tell you - in order for me to do a better job, you need to delete your entire company’s culture and workflow, and have everyone bend over to do things his way, only then can he do reasonable work
oh - and no one tells him how to do his job, NO ONE. he’s always the smartest one in the room and despite doing a terrible job on day one, despite the only results on his resume were vibe coded demos of games that already existed, he demands that you give him a big charter and let him go dark with no communication, taking no feedback
would you have hired a teammate like this?
when humans feel frustrated after using a model, let’s figure out how to RL the model better so they become a better teammate
don’t let a bad model RL _you_
Claude Opus 5 is way better than people give it credit for.
If you're using it like previous Claude models, it's going to suck.
Two things make a huge difference:
- Delete ALL your skills/MCPs/Claude.md/etc. Start fresh.
- Stop telling it how to do the thing. Just say what you want done.
These two small changes make a massive difference.