Sabitlenmiş Tweet
dex
4.7K posts
dex
@dexhorthy
building the post-IDE IDE at https://t.co/hDpglja33W - @aitinkerers sf lead, prev @replicatedhq @SproutSocial @nasa - ai that works pod @ https://t.co/69BhaNtWfd
San Francisco, CA Katılım Ocak 2017
1.6K Takip Edilen14.3K Takipçiler
.@dexhorthy can i re-skin riptide pls
Evan Applegate@youwillmakemaps
"Ghost in the Shell 2: Innocence" had the best burning-amber UIs
English
I don’t disagree wholesale but there’s some shaky points I. Here
you don’t need to run agents in the cloud to publish their progress to the team or make them collaborative
You don’t need agents running overnight to reap the benefits of sandboxes
There are a lot of cloud things that are not containers
English
Running agents locally is a dead end. The future of software development is hundreds of agents running at all times of the day — in response to bug alerts, emails, Slack messages, meetings, and because they were launched by other agents. The only sane way to support this is with cloud containers.
Local agents hit a wall quickly:
• No scale. You can only run as many agents (and copies of your app) as your hardware allows.
• No isolation. Local agents share your filesystem, network, and credentials. One rogue agent can affect everything else.
• No team visibility. Teammates can't see what your agents are doing, review their work, or interact with them.
• No always-on capability. Agents can't respond to signals (alerts, messages, other agents) when your machine is off or asleep.
Cloud agents solve all of these problems. Each agent runs in its own isolated container with its own environment, and they can run 24/7 without depending on any single machine.
This year, every software company will have to make the transition from work happening on developer's local machines from 9am-6pm to work happening in the cloud 24/7 -- or get left behind by companies who do.
English
English
1) not surprising whatsoever
2) this is exactly what I keep saying about models not being powerful enough today
the fact that they can do so much with lossy compression is amazing, but there's no magic here
imo (for transformers) context windows need to be 1-2 orders of magnitude larger for the future people keep saying is reality, and even then the compute is probably not worth it
Lossfunk@lossfunk
🚨 Shocking: Frontier LLMs score 85-95% on standard coding benchmarks. We gave them equivalent problems in languages they couldn't have memorized. They collapsed to 0-11%. Presenting EsoLang-Bench. Accepted to the Logical Reasoning and ICBINB workshops at ICLR 2026 🧵
English
@zeeg @_ChrisCovington @0xblacklight @0xblacklight i changed my mind lets name the code review agent Crammer
English
@_ChrisCovington @0xblacklight @dexhorthy Careful around anyone who names themselves after one of my open source projects 😅
English
the post states “no explanation but this works better than other things we’ve tried”
So on the model side, mostly vibes
I also do think it’s more ergonomic for develops. The xml tags give you a little be more clarity over what’s important to a specific trigger compared to markdown headers - it’s very explicitly where things begin and end
English
@dexhorthy Can you share how you identified/validated this approach? Not doubting, but I want to understand how these evals work in practice
English
Adding opus 1m support meant we revised the context warning in humanlayer from “40%” to “100k tokens”
Matt Pocock@mattpocockuk
Doing some experiments today with Opus 4.6's 1M context window. Trying to push coding sessions deep into what I would consider the 'dumb zone' of SOTA models: >100K tokens. The drop-off in quality is really noticeable. Dumber decisions, worse code, worse instruction-following. Don't treat 1M context window any differently. It's still 100K of smart, and 900K of dumb.
English
Incredibly excited to announce Keycard for Coding Agents - no more copy & pasting credentials or approving individual tool calls.
Agents get task-scoped access, so you can stay in flow and actually build. You’re only pulled in when it matters.
Yolo mode, without compromise.
Keycard@KeycardLabs
Your coding agents inherit your credentials and your permissions. No identity system in the stack can tell the difference between you and the agent acting in your name. Today: Keycard for Coding Agents 🧵
English
this guy gets it
Kent C. Dodds ⚡@kentcdodds
@dexhorthy This is a great talk. Grounded in reality. I'm doing a lot of what you're suggesting naturally (when it matters) and expect my preferred harness (Cursor) to build in some of these features in the workflow (like their Plan mode for example). Thanks for sharing!
English
dex retweetledi
I found this post pretty useful for getting Claude to respect the CLAUDE.md file. Shout out to @thdxr.
dex@dexhorthy
English
@dexhorthy @0xblacklight lowkey kinda want a `source .trashrc` now 😂
English
so you’re saying we should build type annotations for bash!?
dax@thdxr
if you don't have a feedback loop to actually provide type errors it doesn't really benefit yeah the agent can run the type checker manually but it just doesn't it might be over stated though - practically speaking it's mostly just fixing things when it hallucinates an api that isn't there
English
@shanraisshan Any developer on your team should be able to launch Claude in a repo, say “run the tests” and it knows how to run them on the first try, and they pass (and knows how to do any required first time setup on the first try too)
English
@refactorfiend there are product reasons why it behaves this way that I understand and respect. Understeering is better than oversteering in almost all cases. This is an advanced technique for people who know what they're doing.
English
@dexhorthy this feels like it should be a bug ticket that should be submitted to anthropic. or maybe it has and will be addressed two minutes from now in their next update
English
@kentcdodds welcome to x dot come the everything app where everyone is nice and humble and there are no jerks anywhere
English
@dexhorthy Man it's refreshing to talk with someone who has a healthy balance of wisdom and humility (something I sometimes struggle to do myself 😆).
English
Everyone knows that the last 40% of the context window of AI models start to get pretty unreliable...
Except I don't experience this at all 🤔
My two primary AI tools are Cursor (mostly GPT 5.4) and ChatGPT with very long conversations. Cursor compacts and it's still does fine.
English
@kentcdodds (and i think those people who do that tend to feel the most "left behind" by the coding agent madness)
English
thats fair, i did not mean to dismiss at all, but i see how it could come off that way.
I more want to qualify that i'm not expert on every kind of problem and you should ignore my advice if you're not working in the same areas that we're working in
trust me i'm sure nobody prefers spelunking across 6 ten-year-old-repos to get things done 🤣
English