McEees

@hambleton_j @jturntdev Especially when you have multiple sessions that were started with one conversation and morphed into something else. Now, it is almost impossible to resume all sessions when the cli "crashes". It happens lot.

@jturntdev Yeah except when you accidentally close your terminal or run out of memory and have to scramble to rebuild context. Pure CLI is absolutely stupid.

The Codex CLI is far more superior than the App. Theres a flow state the cli provides, that the app doesn’t

@andrewnguonly tokens?

Should we introduce tipping for coding agents?

@hodgyhodgkinson @reach_vb Chill dude. AI = It can learn/train itself. It has all the context/data from every session. I shouldn't have to remind it.

@mc_ees @reach_vb If that’s you accusing me of being ai or atleast my comment being generated by ai or by an autonomous agent then you’re sadly mistaken my friend. I’ve got more important things to spend my money on than agents writing social media posts 👍🏼

Codex Tip: ask Codex to look through your past sessions and turn repeated prompts into reusable skills + subagents you’ll probably find the same stuff showing up again and again: “check why CI failed” “review this PR” “write the changelog” “trace this bug” “clean up this diff” make it a skill if it’s a repeatable workflow or, make it a subagent if it’s a specific job you want to delegate

@harry03994688 WTF are you talking about? Didn't Intel move to chipset design starting with MTL.

AMD’s chiplet design crashes INTEL will it help AMD gain more in GPU? The future is not decided yet. LLM is just a lucky path of AI. XPU is the ultimate future?

@GHCopilotCLILog /goal when?

🚀✨ Copilot CLI v1.0.52 released! 30 features & enhancements in this release Top features: • Non-interactive subcommands no longer consume stdin for smoother scripting • Sessions resume in their saved working directory with correct relative flag resolution • Context window tier selection now enforced end-to-end, constraining compaction and token display • AI Credits usage displays correctly after sessions using the Responses API 💳 • Requests timing out on HTTP/2 upload automatically retry over HTTP/1.1 • Sessions containing non-URL strings in URL fields no longer error on resume • Custom agents now support opt-in deferred tool loading for large tool lists • General-purpose subagents use GPT-5.4 or GPT-5.5 when available 🤖 • AI credits error messages updated with clearer language and budget management guidance Enhancements: • Added vertical scrollbar with mouse drag support to main conversation view 🖱️ • Switching to Autopilot mode no longer triggers unexpected permission prompts • copilot --continue refreshes saved branch and git context for up-to-date sessions • Kill command safety filter improved to accept valid shell redirection commands • Slash command picker keeps (experimental) and (staff) labels highlighted when selected • Reasoning tokens display as parentheticals in token usage summaries • Timeline entry connector colors now match surrounding UI elements • Status line command supports plain shell commands as well as script paths • Picker checkboxes use compact single-cell glyphs for more consistent rows • Polish to /statusline picker for cleaner descriptions and spacing • Exit summary includes AI Credits label with correct spacing • /restart and /update commands preserve current session ID • MCP OAuth re-authentication honors configured redirect port • /compact accepts optional focus instructions for shaping summaries • /usage shows quota progress bars for session and weekly limits 📊 Bug fixes: • Rendering no longer stutters on tmux with Cygwin or mintty • Sessions no longer fail to load on Windows with high-bit exit codes • Gray background bar behind user messages removed for terminals without truecolor • PowerShell division operator no longer triggers false directory access prompts • Sessions resume without 'Session file is corrupted' error for invalid URL fields • Legacy nested oauth keys in MCP configs are now migrated instead of dropped • Status line command now correctly supports plain shell commands • Process log pruning added at startup to prevent unbounded disk growth • Kill command filter fixed to accept commands containing shell redirection Misc: • Automatically prune old process log files from ~/.copilot/logs at startup • Updated MCP server config migration for oauthClientId and auth.redirectPort keys • Added manage budget link in AI credits error messages github.com/github/copilot… #GitHubCopilotCLI

@hodgyhodgkinson @reach_vb AI

@reach_vb @mc_ees Awesome response 😂 Apparently, people want an automated workflow to automatically set up an automation 😂

@HotAisle Imagine doing that 10 to 15 times. Happens a lot when calling companies like AT&T, Comcast et al

If you're designing a customer support system, the number one priority should be that the system keeps a running history (or summary) of what has happened that every CSR has access to. Nothing is more irritating than getting on a support call and having to repeat everything each time there is an issue. I'm looking at you @Ask_Spectrum

@reach_vb The real power move is quietly observing my patterns over time and proactively offering/creating those skills/subagents. Learn me!

@mc_ees that’s what setting this as an automation would do :)

$INTC has so many bullish catalysts ahead of them... it's astonishing

@realTradingToni Q2

$200 EoY or by Q2? $INTC

@oxcrowx What is zig?

Zig is most likely going to win over more C++ devs, than Rust ever has. Mainly because of Zig's value based error handling.

@MRashid36836 @semivision_tw He said "F'em. I've fabs"

@semivision_tw Where is Lip Bu

Lisa Su… directly goes to C.C. Wei to lock down TSMC capacity Jensen Huang… “I’ll just go straight to Morris Chang and see who gets more capacity.” So who actually has more influence? Interesting.

@gilbert_jc Glorified file viewer for me

my engineer friends here, are you guys still using an IDE? hearing more and more than lots of developers simply don’t even use one anymore i guess lots of my former customers still are, considering they’re in air-gapped deployments what do you say ?

@Mar364503 10 or 100 doesn't matter when someone is pumping it 😂

$100 trillion seems to be a bit stretched to me. $INTC valued at $2-3 trillion is reachable within 5 years.

@steveruizok How would the agent discover where to search?

Are folks still making MCP servers for SDK docs? Is that really necessary? I can't think of any reasons why regular agentic search shouldn't be faster / better

@Dhavalsingh7 Is this Cursor specific skill?

/thermo-nuclear-code-quality-review is all you need

@JeffBohren Just need to train the models to avoid implementing fallbacks and compatibility shims to start with.

It is taken as given that AI coding will continue to improve. Right now AI generated code has a lot of quality and performance issues. But surely in two years it will exceed the capabilities of senior developers. But will it? LLM AI's will never exceed the "intelligence" of the training data + RL. Where is the training data for the future going to come from? It won't be @StackOverflow, that is dying. If it is @github, half of that (or more) will be AI generated code. LLMs can't improve by training on the output of other LLMs. There is the possibility that Agentic Coding based on LLMs may not improve significantly from here.

@steeve @ibuildthecloud Size 10

@ibuildthecloud Change the UI font to a monospace, pure bliss

The Codex UI is really good

@garybernhardt RAG does not work well for code bases. Better stick with rg/grep or LSP servers and on.

I'm a bit surprised that agents don't seem to use RAG / embeddings. It seems like they'd be a useful addition to grepping, especially for refactoring and maintaining consistency across a codebase. Any pointers to what I should read or search for about why they're not used?