Reece Hardin

20, Developer @hirealli

[Reece Hardin @TheRealReeceH]

I've shipped over 300+ PRs the past month using @bot and @claudeai Duet.

Duet feels smarter, Grok Bot gets more done without me.

And i'm struggling to choose which one should be my daily driver..

Grok Bot:

Pros:
> Feels like a true EMPLOYEE: I hand it a task and can come back an hour later with finished work. If I had it a single task it rarely needs me to do anything, and me being a single developer @hirealli this really helps how fast we can iterate and improve our products.
> UI/UX: I've truly never used an app that just feels this easy that is this powerful, everything just.. works.
> Speed / Tone: The responses I get actually feel read-able, there not some essay I need to dig through every response, it also just feels so fast.
> X Integration: How easy it is to just work alongside X.

Cons:
> Usage is terrible: I have the Ultra plan, and am scared to ask it to do simple tasks, routines, etc because I can watch my usage go down by the minute.
> Seems less smart: Almost like its effort level or even the model has been changed recently. It used to seem more smart.
> Cursor Agents: simple fix? Cursor agent. simple UI button on a local app? Cursor agent. I've told it to utilize @t3dotcodes and it is a 50/50 on if it launches a new T3 session or new Cursor Agent.
> Poor Tracking: Its hard to know what my agents are working are, where they are in the work, and most times I end up having to nudge and ask.

Duet:

Pros:
> Best Model: Of course Duet's best feature is that it's powered by Opus 5.5 and its so smart.
> Task Management: Very easy to view all my tasks, what is working, what is stuck needing me, and when its completed
> Claude Ecosystem: How easy it is to navigate and steer my claude code sessions and use the overall claude ecosystem straight from Duet

Cons:
> Hand-Holding (MAIN COMPLAINT): I give Duet any task, and it always needs something from me, rather its as simple as a "Yes" on a task i already told it to do, it WILL NOT merge PRs, it can't even merge a PR into my development branch. Which is extremely annoying when trying to have an agent ship real work (i get not shipping straight to a master branch, but even then if I wish to do that, it should..)
> Responses: Every time I ask a simple question, ask it to get something done, it comes back with 3 bubbles I have to read through.
> Tasks/Main Chat: Sometimes it's hard to navigate where it is responding at, sometimes its in the task thread, sometimes it's in the channel, and sometimes it's in both

As always @gabemulley I hope you see this and keep shipping the work on Duet, I see this becoming a main part of my workflow in the next few months.

[Reece Hardin @TheRealReeceH]

🤯Fable 5.2 is actually just insane. The amount of detail these games have is mind blowing.... Let alone the music, sound effects, UI elements, etc...

[Vineer @vineerpasam]

Fable 5.2 is fu*king insane
https://t.co/umU061Ti7A

[Reece Hardin @TheRealReeceH]

Give your agent its own login and testing as if they were a real user. Then have a different model review the code.

When my agents finish something, they log in like a real client, click around, and send me screenshots plus a recording. Then before merging there is a separate model that DID NOT write the code to review and run tests on it.

I've done this and combined it with typical tests and linting, and also utilize the verification skill in PStack from @poteto i've been able to not only ship FASTER but also HIGHER QUALITY code, but also able to ship more accurate code on the first go.

This to me has been really helpful when I launch off agents to get a lot of work done, it's at a point now where it's significantly fewer fixes after the first pass.

[Reece Hardin @TheRealReeceH]

@melvindvivas The issue is none of us truly know what model anything is going to, ideally you wouldn't use Opus for just genuine chat questions, having it use Grok for chatting, then Opus when things get tough would be a great combo in my opinion.

[Reece Hardin @TheRealReeceH]

@CalvinHollywood @bot What subscription do you have, are you running Cursor agents at all? There is a lot of different variables that can drain usage differently.

[Reece Hardin @TheRealReeceH]

So does this come with some kind of reset or we are just stuck waiting a week until our usage resets?? @elonmusk

[Elon Musk @elonmusk]

Just did a round table review with the Grok @Bot engineering team.

Fixes are being implemented overnight and you should notice a dramatic improvement by tomorrow. If not, reply to this post and we will address any remaining issues right away.

[Reece Hardin @TheRealReeceH]

@huesofbanter My agents will use the browser and login with a test login (just like a real client would, then click around). Then it gives me back both screenshots and recordings. I still always test myself before pushing to production. But it’s easy to catch little changes this way.

[Reece Hardin @TheRealReeceH]

I've shipped 300+ PRs in the past month using AI agents, and here's the biggest thing I've learned:

1. Most people are focused on the wrong part.
> They let the AI make a change, then spend their time checking every line before it goes live. At that point, you're the bottleneck. Spend that time writing the verification plan instead. What should the agent click through? What should it test? What screenshots prove it worked?

2. Iterate quickly rollback quickly.
> When letting agents run, letting them run and ship there work is what I see a crucial part of staying efficient, especially when you're a small team. Give your agent clear pass/fail rules. For example: if a fix fails 3 times, stop and ask me. If switching a site over causes more than 20 seconds of downtime, roll it back.

3. Define terms CLEARLY
> For me I made it clear that “Built” isn’t done. For me, a task is done only when it’s live and tested on stg. A lot of agent work gets marked finished as soon as the code is written.

4. Source of truth
> Every agent I use checks Linear before, during, and after working. It logs every change and who approved it, and starts with "X of Y done.", this gets more important as you grow the number of agents you have working on different or complex tasks at any given time.

5. Do NOT stay married to a specific agent/model.
> It's worth taking the time to learn how to build something like a "Bot Relay" where you can use one system (e.g. Grok Bot) then head over to another (e.g. Claude Duet) and build a way for them to communicate back to one another. This way you can get the best of both worlds. (although this needs to be careful consideration, don't just build it to build it and test that new shiny thing. Only build it if you find yourself going back and forth between different AI tools constantly)

There much more, but these are some of the main points.

If you use AI agents I would love to hear more ways I can improve my workflow, so if I missed anything please let me know.

[Reece Hardin @TheRealReeceH]

@melvindvivas There team on Bot said themselves they push new updates to prod every day, they update so fast. If this takes longer then a few days, that shows the priorities on the bot team, this is the one thing all users are complaining about..

[Reece Hardin @TheRealReeceH]

I don’t like this take..

No one says it needs to be free but here is the problem, Bot came out with something that seemed smart, usage was pretty decent.. then overnight WITH NO WARNING, with no offical announcement, it gets completely changed.

Everyone’s workflow they’ve built gets effected. Let’s pretend usage isn’t a problem and never changed, how can you make something just more dumb with no choice of switching to a better model.

Using your analogy, that’s like hiring an employee for $3,000 a month that is extremely smart and gets work done with good quality work and not much hand-holding. Then overnight that same employee gets a significant raise and then forgets how to be so smart and then you have to start hand holding more…

No one minds paying, but we should atleast get decent usage out of it…

[Kyle Reidhead | Milk Road @KyleReidhead]

Everyone complaining about the costs of Grok Bot is ridiculous

It’s providing super intelligence to every human to do work for you

The alternative is to hire humans to do work for you

Many of you don’t, bc it costs a lot and is hard to run a team

Now it’s easy, why should it also be free?

There’s a lot of compute costs to power GrokBot, they can’t just give away intelligence for free

That makes no sense

If you need more usage, pay more, thats how every other product on earth works

[Reece Hardin @TheRealReeceH]

Hopefully this comes out by tomorrow alongside a reset so people can get back to there workflows for the week.

[Elon Musk @elonmusk]

Grok Bot fixes for improved token efficiency and smarter behavior are underway

[Reece Hardin @TheRealReeceH]

@AbeBurnett @bot @claudeai Exactly. And Duet always responds with 3 big bubbles for anything. I can ask a simple question and I’m stuck reading the longest essay lol, Bot just feels like way better to use. I hope they fix the usage and model issue soon, because even with how good it is, its not worth it rn

[Reece Hardin @TheRealReeceH]

I've shipped over 300+ PRs the past month using @bot and @claudeai Duet.

Duet feels smarter, Grok Bot gets more done without me.

And i'm struggling to choose which one should be my daily driver..

Grok Bot:

Pros:
> Feels like a true EMPLOYEE: I hand it a task and can come back an hour later with finished work. If I had it a single task it rarely needs me to do anything, and me being a single developer @hirealli this really helps how fast we can iterate and improve our products.
> UI/UX: I've truly never used an app that just feels this easy that is this powerful, everything just.. works.
> Speed / Tone: The responses I get actually feel read-able, there not some essay I need to dig through every response, it also just feels so fast.
> X Integration: How easy it is to just work alongside X.

Cons:
> Usage is terrible: I have the Ultra plan, and am scared to ask it to do simple tasks, routines, etc because I can watch my usage go down by the minute.
> Seems less smart: Almost like its effort level or even the model has been changed recently. It used to seem more smart.
> Cursor Agents: simple fix? Cursor agent. simple UI button on a local app? Cursor agent. I've told it to utilize @t3dotcodes and it is a 50/50 on if it launches a new T3 session or new Cursor Agent.
> Poor Tracking: Its hard to know what my agents are working are, where they are in the work, and most times I end up having to nudge and ask.

Duet:

Pros:
> Best Model: Of course Duet's best feature is that it's powered by Opus 5.5 and its so smart.
> Task Management: Very easy to view all my tasks, what is working, what is stuck needing me, and when its completed
> Claude Ecosystem: How easy it is to navigate and steer my claude code sessions and use the overall claude ecosystem straight from Duet

Cons:
> Hand-Holding (MAIN COMPLAINT): I give Duet any task, and it always needs something from me, rather its as simple as a "Yes" on a task i already told it to do, it WILL NOT merge PRs, it can't even merge a PR into my development branch. Which is extremely annoying when trying to have an agent ship real work (i get not shipping straight to a master branch, but even then if I wish to do that, it should..)
> Responses: Every time I ask a simple question, ask it to get something done, it comes back with 3 bubbles I have to read through.
> Tasks/Main Chat: Sometimes it's hard to navigate where it is responding at, sometimes its in the task thread, sometimes it's in the channel, and sometimes it's in both

As always @gabemulley I hope you see this and keep shipping the work on Duet, I see this becoming a main part of my workflow in the next few months.

[Reece Hardin @TheRealReeceH]

@Adamdesgns Agreed, but at the same time they are able to ship good quality work fast. Bugs/Performance/Issues has never been an issue with Grok Bot, the usage and model is the only complaint i've seen 95% of everyone having. Just don't know why not they can't do both..

[Reece Hardin @TheRealReeceH]

I feel like if you're at a huge company where every little new iteration needs complete oversight, complete auditing, etc then the current setup makes total sense.

But for people like me who is a extremely small team (just me) it's nice being able to have agents working on small iterations and new features, bugs, etc

It doing that is what allows me to have the time to keep perfecting the verification steps / quality assurance side that way what my agents are working on is all properly aligned with the vision and goals of the company, alongside keeping a good quality of work it does.

[Reece Hardin @TheRealReeceH]

Yes! More useless features that will burn our entire usage in a few prompts...

GrokBot as a software is so well put together (best ease-of-use AI tool i've ever used), but the usage and model behind it really starting to hold it back..

[cradrrat @cradrrat]

Grok Bot’s newest test build, 1.1.0-pre.58, adds six things that aren’t in the current release, 0.79.3.

Hotels - cards show the lowest total price, and a “Check prices” button asks your Bot to show prices for that hotel.

Try-ons - you can keep up to 50 try-ons of each item, and delete one to make another.

X - if your X account isn’t connected, pressing Follow on a profile card asks your Bot to connect it so you can follow.

Teammates - the New Bot menu adds “Create a teammate”.

Tasks - a task box asks “What should a Bot take on?”, and each task shows Working, Replied in chat or Didn’t finish.

Goals - Bots get a Goals panel.

This is all still in testing and may change or not ship. Names, prices and accounts in the images are sample data.

[Reece Hardin @TheRealReeceH]

The competition for Grok Bot is only getting better and better.. meanwhile the @bot team seems to be ignoring what everyone's main complaints and requests are.. what happen?

[OpenBot @OpenBot_]

OpenBot v0.62.0 & v0.63.0 🚀

New integrations

- Generate images, video and audio with Higgsfield.

- Manage tasks and track habits with TickTick.

- Fill saved logins from a dedicated Proton Pass vault.

- Install Hermes Agent directly from provider settings on Mac, Linux or a server.

[Reece Hardin @TheRealReeceH]

@nima_owji What everyone is wanting for Grok Bot: More Usage, Better Model (or atleast more insight on what the models are being routed to)

What Grok Bot team is giving us: Create a teammate button, booking hotels, shopping cards..

It's great a.. optional-feature is great, but come on.

[Reece Hardin @TheRealReeceH]

@ghumanmoiz @poteto Poteto replies to so many posts, yet I haven't seen a SINGLE reply from anyone on the Bot team to a single one of these usage posts (and there is A LOT) so we all know there seeing them..

If there is no fix for usage next week I may cancel.

[Reece Hardin @TheRealReeceH]

@aiclawdius What are you using it for and how many bot/routines are you running?

[Reece Hardin @TheRealReeceH]

I think I may try this, start a new Grok Bot acc, load it with $10 and see how many basic tasks I can give it before the usage is used all the way😂

[Suraj Kadam @thesurajkadam]

i added 5$ credits in my grok bot at November 6, 2026 at 11:45:52 PM PST

and it has been exhausted within 30 mins of work and i had given only ONE TASK which is also incomplete

and now i have given the same chat to claude and told it to continue :)

however amazing the grok bot is, the limits are just stupid!

@bot @elonmusk @X

[Reece Hardin @TheRealReeceH]

@roughneckactual @alexmichael @claudeai @bot @Meta @OpenAI Have you ever used Muse for software engineering? Is it any good at it?

[Reece Hardin @TheRealReeceH]

@aryanvij02 Man your UI looks really good. I followed you and will be keeping a close-eye on Turnstone, that's why I loved GrokBot, how easy all the UI was, easy to send an email, easy to connect plugins, etc.