All the time working with AI, I feel gaslighted. One second it says all good, I fixed everything; a couple of minutes ago, when I asked to verify and found counterexamples, it said, " Oh, my mistake, what was right 100% is absolutely wrong. I'm really tired of doing the same work over and over again because agents are convincing me they fixed everything, but at this stage I basically can't verify it because I absolutely don't understand what's happening in codbase.
1
2
108
This is amazing. Even if the agents did their job, you still need to be capable of understanding the results.
Proud to say that Devin has cracked three more unsolved problems today
1) REFUTED: Graffiti Conjecture 154 (open for ~40 years)
2) PROVED: Graffiti Conjectures 39 & 40 (~40 years)
3) REFUTED: Brandt's Regular Supergraph Problem from West's open problems list (~20 years)
My methodology is explained below, but basically I showed Devin the original tweet and told it to find similar problems and crack them.
The flood gates are open.
Readers added context they thought people might want to know
Graffiti Conjecture 154 was already refuted on June 11, 2026, prior to this post claiming it unsolved. github.com/demonstrandum-…
72
Today I had a dream. Everyone was assigned a totem animal. Some got a sand snake with healing powers. Others got something equally heroic. I remember that the animals were assigned based on your qualities, so I decided to get ahead by acting heroically so I started doing dramatic jumps and sprinting instead of walking.
They gave me a pug.
78
This looks perfect on paper, especially when you are working in the frontier lab and have typed cheat codes for unlimited tokens. With the current token price for frontier models, for levels 3 and 4, it probably makes sense. But it burns money really, really fast. I'm currently at this grade on level 2, but I'm really scared to jump to level 3. Plus, I do believe that proper planning, understanding why and what you wanna build, can beat just running a loop. As soon as the price decreases, I really want to increase my personal level. But now it's kinda not sustainable.
I talk to engineers at other companies every day and hear the same thing: one person is 10x'ing their output with Claude but the rest of the org hasn't caught up.
Watching teams adopt AI, I keep seeing the same 4 steps.
I mapped them out here: Steps of AI Adoption claude.ai/code/artifact/bfdf…
1
109
Looks like a very smart way to save some tokens. Someone encode context in image.
github.com/teamchong/pxpipe
210
It wasn’t intentional, but I’ve reached exactly what the ring planned for me. Trying not to move anymore, to keep it in a current state. Probably once in a lifetime
57
github.com/strukto-ai/mirage
Such stuff start pop up more and more I'm not sure if this gonna be replacement for mcp. Why do we need mcp if we can make everything as a file system
65