Quinn Slack

Btw, for Ghostling (libghostty demo), I didn't write a single line of... anything. Agents wrote 100% of everything you see incl. Nix flakes, CI jobs, etc. I reviewed every line of code manually and constantly nudged the agents in the right direction. I used a mix of Opus+Codex. Note: libghostty itself is of course heavily hand written (with agent assistance all over too, though). I'm just talking about Ghostling itself. Even for CI setup (GitHub actions), I had the agent sit in a `gh` CLI loop watching failures, fixing, pushing, fixing, pushing, etc. Doing GHA with agents are the only way I stay sane, honestly. I just re-read the full main.c from top to bottom and I'm very satisfied. I would've done some things differently but if an engineer I worked with PRed all this, I would've accepted it. Its good enough. Some people have pointed out the commenting is over the top and indicative of AI. This is actually my personal comment style, and I told the agent to comment heavily (its even in the AGENTS.md, look for yourself!). Anyone who has worked with me professionally or in OSS knows that I comment everything all the time.

@Smit_Chaudhary3 @AmpCode Could be a mistake. Sorry! Please email amp-devs@ampcode.com so we can look it up by your email address.

Has @AmpCode free been discontinued outright? I see their note saying they paused it for some "inactive users" but I am far from being inactive when it comes to Amp

@JeopardizeP @AmpCode Sorry about this. Is this happening on every thread? How long has it been happening for?

@sqs @AmpCode I think theres a problem going on with ampcode, it cant create any files and it keeps saying network error and continues to eat up credits. Also love the work you have done so far.

@fire @AmpCode I have gotten one other report of a weirdly stuck Anthropic thread from another customer. Trying to figure out what it is. Will report back if I need more context. Thanks and sorry for the issue here. Let me know if you see this occur elsewhere.

@sqs @AmpCode Yeah fwiw I haven't been able to get this thread to recover, it's just kinda deadlocked. very strange behavior

is @AmpCode down?

This is actually a really positive moment for the acceptance and perception of OSS models.

@TomBishop_91 @AmpCode Sorry about that. It’s likely WorkOS incorrectly thinking you’re a bot. You can email us at amp-devs@ampcode.com with your email address and we can help.

@AmpCode Hi team, I was trying to sign up today and it says "Access blocked, please contact support". are new signups paused?

@swyxio @edwinarbus That is awesome to hear! Yeah a couple months ago someone mentioned you are probably biased and I said I really don’t think so.

@swyxio I thought Cursor had already shipped something like this a few months ago. I’m not a regular enough user to know what specifically is new here. (Maybe they banned your account from accessing this alpha feature? Haha. Idk)

@swyxio It should be there. Cursor.com/glass, then enable, then make sure Cursor is up to date? Took a bit for me to find it too but I was able to try it this morning.

@MonkeyMeaning @AmpCode @claudeai 🫡

switching from @AmpCode to the $200/mo @claudeai plan for a month has taught me an important lesson just use the $200/mo to purchase amp credits

@arvislacis @JebsSteve0x1 @AmpCode Yes, large mode is now Opus 4.6 (also with the same per-token costs for all tokens even above 200k).

@JebsSteve0x1 @sqs @AmpCode I think large mode still uses Sonnet as it's still a lot cheaper than Opus and offers 1M tokens if you really need them. But, anyways, that mode is experimental and only if you really need it.

Smart mode in Amp now lets you use Opus 4.6 wth 300k input tokens of context, up from 168k. Also, tokens above 200k are now charged at the same (not higher) per-token rate. Why not 1M tokens in smart? Quality/cost aren't right. If truly needed, use (hidden) large mode.

Sorry about that. Can you ask Amp and tell it to look in ~/.cache/amp/logs/cli.log to see if there is more information that would explain what is happening? I am not seeing anything on our side. Is it possible you have a packet-scanning firewall? And consistently == 100% of the time, and yet not happening for other threads?

@sqs @AmpCode yeah I have a thread that consistently dies with this error T-019d090f-0a46-76f6-92f6-2d5458021612

@aiDidThisToo @ryancarson @thorstenball @AmpCode We post episodes of ampcode.com/podcast occasionally but aren't doing Build Crew Live anymore.

@ryancarson @thorstenball hi, i wonder what happened on the @AmpCode Build Crew Live. regular podcast on youtube? its my usual go to for advancement of the coding model. its seems dormant. any chance bring it back?

@jetpackjoe_ @AmpCode Like where I’m babysitting it, like UI tweaks or troubleshooting something live.

@jetpackjoe_ @AmpCode Mostly deep lately. At this point we need to rename the modes like smart->interactive. Smart is mostly for interactive tasks for me now.

@jhicks154 Same (usually). Will be in deep mode very soon for everyone.

@sqs prefer gpt 5.4

@pcstyle53 Oh and 95% ish of tokens from an agent are cached.

@pcstyle53 It’s the fundamental architecture of the model. I do think that providers could further lower their cached token prices. Supposedly Claude Max plans don’t even count cached tokens against your quotas. If true, then that indicates there is more room to discount.