I've got an agent in a loop optimizing a renderer with the goal to minimize frame times (and tests to measure). It got times down from 88ms to 2ms and allocations down from ~150K to 500. Sounds good, right? Wrong. This is exactly why agent psychosis is a big fucking problem.
As an experiment, I rewrote the Ghostty core render state in Go, with access to identically laid out data structures as Ghostty and the exact same validation tests. I made a purposely naive renderer (simple, correct, but slow). 88ms per frame with 150,000 allocations (horrendous, lol)!
I then kickstarted a Ralph loop to bring the frame times down. I told it it can't modify input data structures or the public API or tests (they're correct), but it can do anything else it wants. It got to work.
It has worked for about 4 hours. I've spent around $350 on this experiment so far. The results?
88ms => 1.5ms
150K allocs => ~500 allocs
Incredible right? Nope.
My hand-written renderer I ported has frame times (same benchmark) of ~20us (0.020ms) and 0 allocations in the update path.
This is the problem with psychosis and lacking systems understanding. If you don't understand the system, you're going to accept that this is an incredible result. If you understand the system, you'll see better solutions immediately and can do roughly 75x better on throughput.
The people who blindly trust agent output are in the former camp. They're sheeple, overdrinking from a fountain of mediocrity.
Standard disclaimer: I use AI all the time. I like AI. The point I'm making is to not blindly accept results. Think. Analyze. Learn.
I am looking for a killer #flutter engineer. Someone who really masters the inners of it and is obsessed by performance and delightful UIs.
You will be potentially joining a fast growing quick commerce startup in the Middle East and the position is fully remote. #russia#poland
Love this post by @antirez on developing Redis Array support. Its a great showcase of thoughtful AI usage and how AI can empower even the best developers while still producing high quality work. https://t.co/xc5KhcHb2P
10 terminal tools that make you 10x faster in 2026:
1. zoxide
A smarter cd that learns your habits. Type "z proj" and it jumps to the directory you actually meant.
Repo → https://t.co/pZCH3ZQt9F
2. fzf
The fuzzy finder that powers half the terminal world. Search files, processes, git branches, shell history, anything.
Repo → https://t.co/nWEAbBMTm6
3. ripgrep
10x faster than grep. Respects .gitignore by default. Once you use it, you can never go back.
Repo → https://t.co/kb1OcCH9NQ
4. lazygit
Every git command you hate, now one keypress away. Interactive rebase feels like cheating.
Repo → https://t.co/f9fDuU6VKT
5. starship
A shell prompt that shows git status, language versions, and cloud context. Works on every shell. Renders in under 10ms.
Repo → https://t.co/Snotb0pW5B
6. atuin
Replaces your shell history with a searchable SQLite database. Syncs encrypted across every machine you own.
Repo → https://t.co/j0tedZ18wi
7. bat
cat with syntax highlighting, line numbers, and git integration. Your terminal will never look the same.
Repo → https://t.co/omdUbpX14E
8. eza
A modern ls with colors, icons, and git status built in. Makes every directory readable at a glance.
Repo → https://t.co/MxRtY8Jipo
9. yazi
A blazing fast file manager that runs in your terminal. Image previews, async I/O, vim keybindings.
Repo → https://t.co/egfS6pkLfx
10. delta
Turns git diff into something you actually want to read. Side-by-side view, syntax highlighting, line numbers.
Repo → https://t.co/GHJCrGMMm9
@neogoose_btw What about using database for long term storage - and having most of your data in pod memory and read and filter then using roaring bitmaps and things like this ? That’s what we did in a company that process 800k orders per day in the gulf
I never liked databases as an idea.
You literally send a string query over TCP to postgres and it returns you data over tcp as strings.
There is so much potential to make this whole thing better …. but everyone seems to be just fine with it.
Introducing a new cryptographic governance primitive
>> Conviction
- The Formula:
Conviction = Stake x Time.
- Linear Unlocking:
Lock Alpha tokens for a set duration (e.g., 365 days) to prove long-term commitment.
- Mutable Ownership:
If an owner acts maliciously, the community can collectively lock their tokens to a new key and vote them out.
- Delegated Conviction:
Miners and investors can delegate their conviction to cryptographically back teams they trust.
First, testing this primitive on subnets 3, 39, and 81 to help the community regain control before a wider rollout.
* Initial design still being tuned and tested, more information
E071 // Hosted by @const_reborn
00:00 - Addressing the recent subnet owner exploit & subnet governance
07:18 - Introducing the "Conviction" mechanism
11:38 - Rollout Plan: starting with subnets 3, 39 & 81
18:42 - AMA Begins
21:03 - How subnet teams can still raise capital (OTC)
24:47 - Q&A: Hostile takeovers & malicious whales
41:22 - Why centralized competitors can't buy and kill decentralized subnets
54:14 - @mogmachine joins: Scenario testing subnet impacts
Drop your thoughts or questions ↓
#Bittensor #TAO #CryptoGovernance
@halhjri14 AI is not magical. AI generates code but it has never been the bottleneck.
These Saas, as bad as they might be, still made it to convince a market to pay and use so they probably have skills that make them valuable.
2 smart AI engineers won’t build you a working business.
anthropic had mythos internally since 2026-02-24, so the leaked claude code (2026-03-31) was being written by it for over a month. if you checked the code, knowing these pieces of information will help you temper your expectations.
Adam Back, résident maltais, inventeur de la preuve de travail du protocole bitcoin, était Satoshi Nakamoto selon John McAfee
Le New York Times reprend cette conclusion sans la citer.
Pour ma part je continue de penser que c’était Hal Finney, parce que seul un mort aurait pu faire x1000 000 sans jamais vendre…
(Et bonus : cette image 👇)