Phil Stőck

Any page. One click. Clean context. PagePie turns messy webpages into Markdown your AI actually understands. Claude, ChatGPT, Grok, Gemini, local. The crumbs stay on your device. Waitlist: pagepie.app

Touching grass? 😂

If the pace of cybersecurity events persist, I might burn all my @bot tokens just to keep up with the threats 😂

I have a feeling that Service Canada's online services have been compromised. I just received an email for my unemployment T4 document for my income tax, which is weird to receive at this time of the year. Might be legitimate, but I won't open it and wait till next January to receive a legitimate one

I wonder if an AI agent is behind France's fridge hack 🤔

That's BOLD

Apparently, there are people being outraged by the torture chamber experiments performed on a Qwen model this week. The Free Britney movement has gone numeric at once.

I've got inspired by @LegalPrimes visualization, and wondered how we could step it up with a glimpse on the 5th dimension. The idea would be that a change in the view angle relative to the prism would reveal another probable future.

I had it in reverse, and that's why my results sucked. For a while I tried to make Grok Bot the place software gets built. Specialized coding bots. Skills and rules mirrored into Cursor. pstack concepts on paper. The goods still came back half baked. Then I flipped it. Cursor owns the how. Cloud Agent. pstack. Implement lives there. Grok Bot stays the outer loop: what, and done. After the coding seats came out, a local toy that used to stall came back clean. UI. Favicon. First run that actually felt finished. Same tools. Wrong seat for the craft. xitter.gq/PhilShteuck/status/210…

So far, I can't get Grok @bot solve software decently. I have some specialized bots, implemented pstack concepts, hired Dr. Eggbot, set the skills and rules up on Cursor... still, I only get half baked goods. I take the same project, build it with Grok Build, and have a decent results with the first try. I have a hard time figuring out what could be the issue... something seems to be lost with the handoff from/to Cursor. Or is it Grok @bot harness? 🤔

Unless the robot has some kind of sensors the boxer can aim at in order to disable the robot... this combat is non sense.

I keep two lists. The first one is the job. Companies that actually fit. Size. Reality. Who answers. That is where the week goes. The second list sits beside it. I call that the Dream 100. Bigger shops. They feel off-limits. Harder to get into. Better volume if you ever got in. Better endgame. More notoriety. I would not dare calling them first. You still work the fit list. The Dream 100 is the scary pile you keep in view, not the main job with a cooler name. Which list are you actually working this week: the fit list, or the scary parallel pile?

This morning’s pickleball tournament confirmed it: my soft game is elite, and my serve has trust issues. Missed most of my serves. Not because I forgot how. Because I really wanted to not miss them. Anxiety is a fantastic coach if your goal is creative ways to find the net. Dinks? Smooth. Head? Locked itself out of the car. Next reps are looseness, not more technique. The paddle was fine. The operating system needed a reboot.

I work in Grok Bot with a head of staff. The day's tasks used to go into chat. The whole list. Paste. Lose the thread. Paste again. A mess. So we made shared folders. Inbox is intake. Outbox is where the results come back. Less pinging head of staff for every scrap. During the day I collect a todo list. I drop it in the inbox. An automation plows through that list a few times a day. I am not sitting there babysitting the paste box. When a task is done, it lands in the outbox. Where do you put the work when chatting it all becomes the bottleneck?

First live pitch practice. I thought I would say the thing, take a couple of light no's, keep going. The rejection hit harder than that. One wrong word. A mishear. The call went sideways. I almost hung up. Felt useless for a minute. Same body-feel as early pickleball serves. I was so bad I wanted to leave the court. I stayed. Serve is decent now. Not pretty. Decent. Uncomfortable now is the same engine as a hard engineering problem. You stay on the court. Where are you tempted to hang up that you should stay?

On another project I set Claude Code up more like a factory. Architect. Implementer. Tester. Then harsh review, pstack-shaped, the kind of pass that is allowed to say no. What showed up were edge cases a normal QA pass would miss. Small. Easy to wave off. The kind that become a late night later. It burns tokens faster. That part is honest. Not using it still costs more, especially against a late refactor, if a refactor is even possible. The other payoff: enough trust to deploy more agents instead of babysitting every diff. Burn tokens on review now, or on surprise bugs later?

When you switch back to Opus 4.7 because your done with Opus 5 and Fable 5.1 being sorry for messing up your architecture ! 😭

Two weeks of testing pickleball paddles. Different feels. The shiny ones got a turn. The quiet result: the regular paddle is still the one I prefer, and the one I play best with. Not a product review. A small human experiment. Shiny lost to the boring one that already works in my hand. What shiny tool lost to your boring default?

Started with Grok Bot like vanilla chat. Talk to it. Hope it remembers the bar next time. Then I put skills and guardrails in. Not to go faster. So I can delegate slowly, with more confidence. The bot has a written bar instead of a vibe I re-explain every session. It costs time up front. Usage too. You shape the skill incrementally toward the result you actually want. Later, when you need that job again, it already matches your criteria. Sibling to yesterday: showing the outcome is the ask. Skills are how that bar survives the next chat. What is the first skill you would write so you stop re-explaining the same bar?

I used to chat the bot like a play-by-play. Every step. Here is what I am doing. Here is the next click. Here is why I am thinking this. This week I moved that time. Less narration. More showing the done. The ask is the outcome I want to see, not a tour of my thinking. Bots land closer when you show the finish, not the walkthrough. You still have to think. You just stop making the chat the thinking. Less time narrating the steps. More time showing the outcome. Are you narrating the steps, or showing the done?

You don’t need one marketing bot. You need jobs with clear handoffs. Same assembly-line idea, different department. Head Marketing sharpens the playbooks and critiques. Line marketers own the brief. Copy inherits the brief and writes the final sentences. Marketing does not write the post. Copy does not invent the strategy. When the seats have names that mean a job, the tool stops being decoration. It starts being leverage. Costume names feel fun until nobody knows who owns the next move. If you are adding AI seats this week, give them a handoff, not a vibe.