{ nestedlogic }

👨🏻‍💻Dev w/ a decade of questionable experience | Devoted husband & dad to 4 beautiful boys | Building TinyTeams, a spiritual successor to SmallBall

[{ nestedlogic } @nestedlogic]

Shipped alpha! Have fun y’all: https://t.co/hFkkrWcdpB

[{ nestedlogic } @nestedlogic]

I haven’t run any benchmarks, but my primary workflow for driving a project, a project being a subset of issues oriented towards a particular milestone/vertical slice, is that each task or issue or story begins with plan mode. My motivation is that “plan mode” contains an opinionated prompt or guardrails built by the harness team that improves the outcome of the execution. It also provides an obvious touchpoint for the operator.

All of this can be built into the skills that drive my workflow, but I haven’t done the work to design my own “plan mode.”

[{ nestedlogic } @nestedlogic]

@dhh Do you find the same philosophy extends to the digital workspace? Insofar as it's meant to clear the mind.

[{ nestedlogic } @nestedlogic]

@nateberkopec I recently switched from Astra / medium to Astra / xhigh (after the astra / medium FUD) now down to sol / xhigh for worker agents. In Claude's ecosystem I prefer Opus / medium (the frontier code benchmark is stuck in my brain).

Why sol / medium, if you don't mind me asking?

[{ nestedlogic } @nestedlogic]

I saw that! And I plan on giving it a go, but I use Claude Code as well and enjoy using certain models across different providers for certain tasks. I've been using herdr as my control plane, and while herdr is great as a TUI, after using the ChatGPT app exclusively for some recent projects, there's something really appealing about an opinionated UI across devices - although my workflow hasn't translated well to it (inter-agent messaging, orchestration, etc).

All that is to say, I'd be interested in hearing what sort of financial pressure you feel using API pricing (or whatever discounts you're able to provide through AMP) versus the subsidized pricing.

[{ nestedlogic } @nestedlogic]

@gladimdim Not at all what I was implying.

[{ nestedlogic } @nestedlogic]

@VictorTaelin Meanwhile my Astra has been working for like 24 hours on a WAL-based push system and just can't seem to crack it.

[{ nestedlogic } @nestedlogic]

OpenAI needs a Claude Design equivalent.

[{ nestedlogic } @nestedlogic]

@RhysSullivan Try $400 a month because you can't commit to one or the other.

[{ nestedlogic } @nestedlogic]

I'm finding the ChatGPT / Codex distinction poorly legible and unwieldy in-app. For instance, I might be working in Codex on a local project and want to ask ChatGPT something unrelated (in the cloud). What I find myself doing is switching from Codex to ChatGPT, which preserves my running Codex conversation - convenient but confusing; then, start my conversation.

[{ nestedlogic } @nestedlogic]

@thsottiaux first, appreciate the tools you and your team are building. I’ve been in the terminal exclusively but decided to give the ChatGPT app a try for a new project and am really enjoying myself.

Second: I’m not sure how this is possible but I somehow managed to tank my weekly usage in under an hour. It could be some of the orchestration experiments I’ve been running, to replicate what I can do with herdr, or something else? Highest effort used was astra / xhigh (no fast mode).

If it’s interesting for the team to take a look let me know and I can get you whatever diagnostics you need.

[{ nestedlogic } @nestedlogic]

I started a ChatGPT "Work" conversation on my desktop to make some headway on our upcoming vacation, which turned into a "remote" (local to my pc) conversation which was unable to use Instacart for some reason, so I had to ask the agent to transfer the conversation to ChatGPT Work in the "cloud."

In its defense it was able to transfer it and got me into the right state, but I'm still confused as to how I should have begun the process.

[{ nestedlogic } @nestedlogic]

What are some habits that work well with the Codex Desktop app? I've been living in the terminal for a long time (herdr) and I want to test drive something new. So far so good but I'm getting context anxiety not seeing the auto-compaction run like it does in the Codex CLI (or I'm missing it). What's the consensus on long threads? Any other best practices I should know about?

[{ nestedlogic } @nestedlogic]

@omair_inam @Shpigford That is correct, custom sub-agents can be pinned to a specific model in their front-matter, but built-in ones (Explore, Plan, general-purpose) inherit from the main thread. That's detailed in the same documentation you just linked.

[{ nestedlogic } @nestedlogic]

@Shpigford Medium for all tasks, even planning / ideating? The limits are brutal and I’m trying to find ways to manage. For instance, I have a general purpose implementation agent that runs Opus / Medium that I let Fable / High manage. Especially after seeing this.

[{ nestedlogic } @nestedlogic]

@wesbos Our two year old recently smashed our 75" with a mug; it was an expensive TV, probably ~$2,500 at the time of purchase. Didn't make sense to invest that kind of money (again) while he's still in his "throwing stuff" phase so we opted for a mid-range TCL for about $1,000.

[{ nestedlogic } @nestedlogic]

@dhh My man.

[Tiny Teams Baseball @TinyTeamsDev]

Insane web gem from the third baseman. This is a preview of some of the work we've been doing to make fielding more dynamic.