Polkbh

using #ChatGPT , Claude, and Gemini for everyday work stuff

sharing the prompts and shortcuts i keep coming back to

[Polkbh @Polkbhtx79]

Making one video with AI is easy. Making a consistent series is harder.

Here’s a practical workflow for creating videos with AI and code—from the first brief to the final review. https://t.co/kUOzfX6ATr

[Polkbh @Polkbhtx79]

@elonmusk @bot Grok Bot’s helped me sort out a lot of stuff.

[Polkbh @Polkbhtx79]

@bot Having the same Bot in the meeting and doing the follow-up makes a lot of sense.

Link each task to the decision behind it, so when the team changes direction, the Bot can flag what needs updating. That could save a lot of rework
https://t.co/C7vMSpL1kA

[Polkbh @Polkbhtx79]

Grok Bot 1.0 is starting to look like a real coworker.

With Grok @bot 1.0, you can bring it into a meeting, discuss the project together, and give it work to handle afterward.

When you assign that work, it can draw on the conversation it just heard. That could save the team a lot of explaining the same background again.

The launch and recent updates make a few team workflows worth trying

1. Share one Bot with the team
A Team Bot can work with specific people or groups in Slack and Microsoft Teams, so everyone can turn to the same Bot.

As teammates teach it how they work, suggested changes to shared knowledge and skills go to the Bot’s owner or a Manager for review. Team Bot sharing is available on Team and Enterprise plans.

2. Bring it into meetings
That shared Bot can also follow the project through meetings. It can join Google Meet, Zoom, or Microsoft Teams, take notes, answer questions in the meeting chat, and send a summary and action items afterward.

Send it the meeting link, approve the request, and have the host let it in. Meetings are available on Team, Enterprise, and Ultra plans.

3. Give it a clear role in the follow-up
What interests me is how this could fit into a weekly project meeting.

Ask the Bot to record each agreed task, who owns it, when it’s due, and which decision it came from. Then give it a specific follow-up task, such as gathering research or drafting a document. Before the next meeting, have it check progress and flag anything that needs a decision.

Keeping the original decision attached to each task would be especially useful when plans change. If the team changes the audience for a launch, for example, ask the Bot to find the copy, slides, and research tasks that need updating. That could help catch work still following the old plan.

4. Let it suggest what needs attention
Grok is also emphasizing more proactive behavior, including spotting unfinished work and offering to help.

The usage guide describes this as proposing work first and proceeding after approval. A clear assignment and follow-up schedule would give the Bot something concrete to work from.

5. Connect tools and follow task progress
A couple of recent updates could help with the execution side, too.
- Tailscale support lets Bots reach MCP servers on a personal private network, giving people with their own tools a way to connect them.

- Cloud Agent conversations and task states are visible in the chat and Tasks panel, making delegated work easier to follow.

For a team already using agents, a recurring project meeting seems like a practical place to start. Give the Bot a defined part of the follow-up, then see whether the next meeting takes less preparation and whether the work reflects what the team agreed on.

[Polkbh @Polkbhtx79]

Grok Bot 1.0 is starting to look like a real coworker.

With Grok @bot 1.0, you can bring it into a meeting, discuss the project together, and give it work to handle afterward.

When you assign that work, it can draw on the conversation it just heard. That could save the team a lot of explaining the same background again.

The launch and recent updates make a few team workflows worth trying

1. Share one Bot with the team
A Team Bot can work with specific people or groups in Slack and Microsoft Teams, so everyone can turn to the same Bot.

As teammates teach it how they work, suggested changes to shared knowledge and skills go to the Bot’s owner or a Manager for review. Team Bot sharing is available on Team and Enterprise plans.

2. Bring it into meetings
That shared Bot can also follow the project through meetings. It can join Google Meet, Zoom, or Microsoft Teams, take notes, answer questions in the meeting chat, and send a summary and action items afterward.

Send it the meeting link, approve the request, and have the host let it in. Meetings are available on Team, Enterprise, and Ultra plans.

3. Give it a clear role in the follow-up
What interests me is how this could fit into a weekly project meeting.

Ask the Bot to record each agreed task, who owns it, when it’s due, and which decision it came from. Then give it a specific follow-up task, such as gathering research or drafting a document. Before the next meeting, have it check progress and flag anything that needs a decision.

Keeping the original decision attached to each task would be especially useful when plans change. If the team changes the audience for a launch, for example, ask the Bot to find the copy, slides, and research tasks that need updating. That could help catch work still following the old plan.

4. Let it suggest what needs attention
Grok is also emphasizing more proactive behavior, including spotting unfinished work and offering to help.

The usage guide describes this as proposing work first and proceeding after approval. A clear assignment and follow-up schedule would give the Bot something concrete to work from.

5. Connect tools and follow task progress
A couple of recent updates could help with the execution side, too.
- Tailscale support lets Bots reach MCP servers on a personal private network, giving people with their own tools a way to connect them.

- Cloud Agent conversations and task states are visible in the chat and Tasks panel, making delegated work easier to follow.

For a team already using agents, a recurring project meeting seems like a practical place to start. Give the Bot a defined part of the follow-up, then see whether the next meeting takes less preparation and whether the work reflects what the team agreed on.

[Polkbh @Polkbhtx79]

Is this how your Grok @bot is burning through its quota?

Set it to “check every 15 minutes” and you’ve scheduled 672 wake-ups a week. Even when nothing has changed.

Each time, it still has context to read and steps to work through. If other bots join in, their replies add to the usage too. That’s how the paper “Grok Bot Token Usage” breaks it down.

So where can you cut back?

1. Start with anything running on a timer. A Slack mention, an email, a GitHub event, or a webhook can trigger the task when something actually happens. Test runs count too. In the paper’s worked example, cutting wake-ups alone roughly halved usage.

2. If a bot has been working in the same conversation for six weeks, it’s carrying a lot of old context. Write a short handoff, around 300 words, and move the work to a fresh bot. Trim its profile and memory while you’re there. Those get read at every step too.

3. Tell it when to stop. Define what counts as done and add a rule like “ask for help after two failed attempts.” Otherwise, retries can keep eating into the quota. Connected apps can save steps here too, cutting out the need to open pages, take screenshots, scroll, and click.

4. Pick one bot for each job. Four bots in a group chat can turn one question into four billable replies. Put one bot in charge and cap bot-to-bot exchanges at two messages.

5. Already at 80% halfway through the week? Pause lower-priority routines and ask for just the final result. The write-up says Grok chat, Grok Bot, and Cursor share one weekly pool, with no per-bot breakdown or warning before it runs out. It also recommends setting a spending cap before you hit the limit.

The paper’s worked example brings usage down from 2,000 units to 400, or 100% to 20%. Your savings will depend on your setup, but the changes are straightforward.

Let events trigger the work, keep the context lean, give retries a limit, and put one bot in charge of each task. Once those are in place, see how much of your weekly quota the same work takes.

[Polkbh @Polkbhtx79]

Having the plan, tracker and data in Claude could save a lot of jumping between apps. Less copying things around would already make this useful for me.

[Claude @claudeai]

Starting to roll out today: Claude Whiteboards, Claude Sheets and Claude Data.

Think a plan through on a board, build a tracker in a sheet, and tell a story with your data.

[Polkbh @Polkbhtx79]

@C1234r23 Planning how the ending loops back to the start makes it so much easier.

[Polkbh @Polkbhtx79]

GPT Image 2.6 can turn a still image into a short AI video. How far can you take it with a five-stage storyboard?

Here’s the workflow behind these four loops, with a reusable prompt and a few lessons from the results. https://t.co/2IoM4HuW1o

[Polkbh @Polkbhtx79]

@distortgeekin Feels like having a personal assistant now

[Polkbh @Polkbhtx79]

@bot Time to resume all those paused tasks.

[Polkbh @Polkbhtx79]

Worth saving if Grok Bot is part of your daily work.

[Marko @Marko_Poly]

https://t.co/G0F7SXoClD

[Polkbh @Polkbhtx79]

@hqmank Both would make the wait worth it lol

[Polkbh @Polkbhtx79]

@MrTacticalX The next model needs to save enough time on corrections to make switching back worth it. Extra usage alone wouldn’t do that for me either.

[Polkbh @Polkbhtx79]

@tivollem This is really helpful.

[Polkbh @Polkbhtx79]

@Jtmatio Does the sampling stage after upscaling preserve the original motion, or does it start changing things?

[Polkbh @Polkbhtx79]

Your next post idea could be in a reply you missed or an article you saved.

A practical guide to using Grok Bot for topic research, weekly reviews, and finding ideas in your own material.

Copy-ready prompts included 👇 https://t.co/92SxSUjwSo

[Polkbh @Polkbhtx79]

Prompt : Organize my AI learning notes into an index I can look up whenever I need a method.

1. Write everything in English.

2. Use a table with three columns:
When I need to… | Use this | Have ready

3. Describe everyday situations in plain language. Put one method in each row. Split combined notes and merge duplicates while keeping useful differences.

4. Use only my notes. Don’t invent features, steps, or results. Mark missing details as TBD and ask short questions to fill those gaps.

5. Below the table, keep the steps, prompts, and reminders from my notes, grouped by method.

6. Save the index and supporting details to AI-Learning-Index.md in this project. If the file already exists, update it while preserving unrelated entries. Tell me where you saved it.

7. When I ask you to look up a task in this project, read AI-Learning-Index.md and find a relevant method. If nothing matches, say so.

Keep it concise. No introduction or summary.

My notes:

[PASTE NOTES HERE]

[Polkbh @Polkbhtx79]

I've been learning AI shortcuts, then forgetting them when I need them.

Then I realized I could just paste my notes into Codex and ask it to organize them by when I'd use them.

Now I can find a method when I need it.

Full prompt in the replies.

[Polkbh @Polkbhtx79]

@dextune This reminds me of the little games I play on Steam

[Polkbh @Polkbhtx79]

Building a website with AI gets easier when you have a clear design direction.

A reference gives you something concrete to work toward: a layout, a type treatment, or an interaction you can describe to your AI tool.

These 8 resources can help:

1. MotionSites Studio
Start with a design, then use AI to customize it and generate the code. Useful when you have an idea but need a visual starting point.
https://t.co/gAVKE8fb2u

2. MotionSites
Website prompts for 3D designs, animated sections, and hero layouts, ready to use with Claude, Cursor, or Lovable. A place to start when you're struggling to describe the look you want.
https://t.co/Mao5jKJQlE

3. Land-book
A curated website gallery with clear categories. Use it to compare layouts, study content hierarchy, and choose a direction before you start building.
https://t.co/yre9pMObfy

4. Landdding
A collection focused on landing pages. Useful for studying how a page introduces a product, presents its benefits, and guides visitors toward the next step.
https://t.co/GgLXgxKS7f

5. Recent Design
A gallery of recently launched websites. Browse it for fresh references when you want to look beyond the examples you keep seeing.
https://t.co/W2zikxfFNE

6. 60fps
Mobile UI animations and microinteractions. Use it to think through transitions, feedback, and how an interface should respond to a tap or swipe.
https://t.co/1tCv0DKgwj

7. Font In Logo
See which fonts brands use in their logos. A starting point for exploring typefaces and the visual character they bring to a brand.
https://t.co/o2Q9yka9X4

8. Godly
A curated gallery for exploring distinctive web design. Useful for studying composition, typography, and visual details that you might adapt to your own project.
https://t.co/0O89WvhlEE

A practical way to use this list:

Pick two or three references that fit your project. Note exactly what you like about each: the spacing, navigation, headline treatment, or interaction.

Give those observations to your AI tool alongside your content and requirements. Build the first version, then refine the typography, mobile layout, and motion.

The more specific your references, the easier it is to explain what you want.

[Polkbh @Polkbhtx79]

Github: https://t.co/4dbMTs4VkV

[Polkbh @Polkbhtx79]

I came across vibe-knowledge recently and started experimenting with it in Codex.

It’s a skill for making short knowledge videos, with 44 visual styles to work from. I wanted to see how those styles would handle different subjects, so I picked nine and wrote a separate prompt for each.

I tried DNS, backups, caching, meeting notes, desk organization, and a few other everyday topics.

My first batch looked pretty similar, so I went back to the style descriptions and thought more carefully about how each idea should look. A blueprint made sense for a DNS lookup. A pixel game gave the backup rule a simple progression. For caching, I used an isometric model to show a data packet taking a shorter route.

That revision also changed how I wrote the prompts. Instead of leaving it at “explain caching,” I described the action—a request reaches a nearby cache, while a longer route leads to the origin server. Giving Codex something specific to animate made it easier to judge whether the scene was doing its job.

Once the individual scenes were ready, I put all nine into a grid. They play together, then each expands to full screen for its explanation before returning to its original spot.

That’s the video below.

Putting them together exposed a few things I’d missed. Some labels overlapped, a few lines of text were too small, and I checked every zoom to make sure the panels returned to the right places. Those adjustments were part of getting the video ready to share.

I built the animations with Codex, following the skill’s style directions, and used HyperFrames for the final render. I’m still exploring the full production workflow, but this experiment gave me nine examples and a set of prompts I can build on.

That’s what makes a tool worth keeping around for me—having something useful to start from the next time a similar task comes up.

I’ve included all nine prompts if you’d like to adapt them to your own topics.