J.

i am meat popsicle.

[J. @atlassianengine]

it only really works when every site just allows agents to make reservations via webmcp and not computer use only.

when it can go and find and source manufacturers and deal with entire conversational flows, signups and communication, maybe it's useful then.

the ecosystem isn't there yet.

and maybe it never will be.

[J. @atlassianengine]

Openai codex plugins and app oauth have opened the floodgates.

If anthropic/grok don't reply in kind soon, they lose the ecosystem war.

Apple didn't win on better phones in the longterm. it won because of the app store.

Grokbot integrations lead on the frontier side, which OpenAI have now almost identically implemented, but that doesn't give the app inside Grok directly.

I'm surprised if it won't be long when openai starts charging a percentage for subs sold from within ChatGPT.

Looks to play out that way rn considering they explicitely state it goes against the app rules rn to include links for signup, payments and so on.

I imagine they're still building the architecture for it and will roll it out within 3 months.

My assumption is they're testing uptake and integration, before eventually the start tacking on that 10-30% 'app store' fee that allows Sign up links to go through their routes inside the app first.

Well. Of course, this is also dependant on traditional companies surviving long enough to keep apps in such way. In fact, it might be their only lifeline for some of them.

I've had custom apps built explicitely for myself, running in my own harness, for almost a year now.

People are already replacing paid solutions with their own, that work within their own ecosystem.

I also don't need the integrations myself. I built them already, completely native to my ecosystem. Who the fuck needs a 'Todoist' inside of Codex when it takes 2 days to build one from scratch. 'Clockify', forget it. 3 hours - done. Mine goes further than that, but this isn't the post for it.

Codex doesn't solve the issue entirely. There's still no memory substrate behind it and the schema's all mixed up. Your Figma board still can't talk to that to-do list extension you added.

And nobodies going to want to keep paying 100x subs for features and mini saas apps.

Though I do see it as an opportunity to embed some of the things I've been doing in my own harness and building within the ecosytems that allow me to.

So please @AnthropicAI and/or @grok, allow me to also grace your presence with native Oauth inside my own app, along with app embeds.

Just a thought.

[J. @atlassianengine]

why hasn't @grok @SpaceXAI literally just plopped their web app into a tauri/electron app yet and released a surface people actually want to use?

Even if it's not perfect right now, it'd take 60 minutes to wire in the Oauth and functions from CLI, and get it out and build on it from there.

[J. @atlassianengine]

@luckeyfaraday Impressive. Let's see 5.5's Mario.

[J. @atlassianengine]

@NovaXCode Try going barely any sleep, 3-4amers, no gym and sat on a desk talking to Codex for 18 hours a day.

You'd look the same in 12 months.

[J. @atlassianengine]

The local Worker loop is useful.

What I dislike is how hard it is to get a trustworthy, disposable copy of our whole application before merging. I run several Workers with local state and service bindings, so I maintain custom orchestration and configuration checks locally, then use a separate governed staging release to prove the real Cloudflare setup.

What frustrates me is the “deploy, discover missing Cloudflare setup, configure it, redeploy” loop. The Worker can be ready locally while account-level resources, hostnames, or security settings still need to be created and wired up before the real integration works.

I've started moving that work earlier.

Your recent Turnstile change was a big plus. Previously it had to be setup post deploy. Now, my locally run control plane can provision a Turnstile widget before it's live, and securely carry its generated credentials into the Worker release.

We’ve also had to configure and verify staging-side settings like R2 CORS and Email Routing.

I’d like Cloudflare’s tooling to make that pre-deploy workflow first-class: plan changes to environment resources, apply and read them back before releasing the Worker, and clearly flag anything that still needs a true edge test.

That way, staging is where we verify the release; not where we first find out what Cloudflare setup is missing.

[J. @atlassianengine]

@tobiadonadon_ reviews.

[J. @atlassianengine]

the bell curve once you've built from scratch your own systems is phenominal.

if you build from first principles, and aim for package reuse, at some point you ship 6 months of features in 6 hours.

and it will just keep compounding.

a day for me looks like a year for somebody else.

that is what you're competing with.

and soon they'll all be on 10x $500 plans whilst you're on a $100.

this is what the fear of missing out is. this is why everybodies going fucking nuts on 'winning'.

it's simply being left behind. ai will enhance everybody, but at some point it's impossible to catch up.

[J. @atlassianengine]

nobody actually cares.

they'll pay it, because that's the cost to win. openai already know the cost to not have it far outweighs the $500 price tag for mega users.

everybodies competing for something, but we've no idea yet what that 'winning' looks like and how things change in 5 years.

[J. @atlassianengine]

@thsottiaux where's the link on how to set it up 'officially', so we're not riding on the 'unofficial' route we were going before.

[J. @atlassianengine]

@thsottiaux @grok how much is decisions api cost vs jev api cost?

[J. @atlassianengine]

sonnet 5.5 high usage is definitely broken.

medium feels like it barely moves at all. it's like luna level usage but writes 100x faster.

then you switch to high and it moves it more than it does opus.

[J. @atlassianengine]

@arampell hello

[J. @atlassianengine]

@JustLingonberry makes absolutely no difference whatsoever.

They'll still find someway to make it insufferable to use.

until they fire the entire antigravity management team and replace them with anybody competent, then there's no saving the perceptions of gemini.

[J. @atlassianengine]

@ChadAppDev x is for posting shovels and not telling people where you found the gold.

[J. @atlassianengine]

tbf,

the fact that openai support full custom harness production and allow you to basically do anything with the sub (minus subtoapi of course) is a gift.

so i give them props for that; and am incredibly grateful i can implement that fully, with websockets and everything else, right into my own custom harness.

it gives you so much more power than brute forcing your way in via varying cli methods.

[J. @atlassianengine]

the only thing in my opinion truly left instantly unvibecodable is outreach.

think stategic reddit posting, cold emails, posting from fake accounts that look real on facebook.

that requires time, management, and build up over a few months at the very least from a network of warmed accounts preferably running from different IP's.

which is why nobodies truly actually building 'marketing apps'.

cold calling systems are closer to vibecodeable but require an outreach team unless you're happy to use ai voice bots. social schedulers are easy. videos/ads are even easier with the right system.

it's not impossible. if others have done it, that means it's possible. but it's a 6 months to a year wait, especially for email, before you can truly utilize it.

but like a lot of things in saas, it's one of those, pay for it now whilst it's easy, then build your own over time that slots right into your existing system.

at least it is with my mindset.