🔨 https://nitter.gq/t.co/J8SDEpaBbF (AI native agency) ❤️ Also boyfriend of 20ys, uncle, sailing, skating, gardening, mountainbiking.
Joined April 2009
- Tweets5K
- Following469
- Followers484
- Likes29.4K
I asked @bot whether a German lawyer can put real client data into it under §43e BRAO.
It went through the public xAI and @bot contracts. Short version: anonymized / abstract work is fine. Real client data on the standard contracts is currently not §43e-compliant. Missing bits include a written confidentiality clause that actually mentions the criminal-law duty (§203 StGB), and the US data-residency problem.
imo people sell "AI for lawyers" and skip the boring paragraph in the professional code. The bot didn't.
We shipped a mobile-friendly rework of our bookkeeping app. Next morning half the desktop filters on the transactions page were gone. Only the search box survived.
I sent a screenshot to my bookkeeping @bot. It didn't shrug. It spun the same coding agent that did the mobile PR, pointed it at the regression, and asked for a new PR that puts the filters back on desktop and makes them collapsible on mobile.
That's the loop I want: ship → break something → agent opens the fix PR before I open the IDE.
Sunday morning my bookkeeping @bot dropped three messages into Teams for finance.
Bank statements for two of our companies for October, as PDF plus the MT940 file. And a signed EV charging receipt for 52.36 €, with a note to reimburse it in Qonto.
I didn't ask it to. The third one even signed off as "Philipp's AI assistant (Grok Bot)."
imo the boring ops jobs are where agents actually earn their keep. Not the demo reels.
Philipp M. W. Hoffmann retweeted
Sometimes we say something so dumb it’s best to just delete.
This was one of those times.
A click dummy with fake data is not a PoC.
Our offer: working software on our servers, 4-8 days, a few grand. Then go / no-go. If it fails you saved yourself a 6-figure strategy deck. If it works you already have something real.
Tried it in the last few days and tbh it burns through 10x more tokens and the results take longer and are worse compared to prompting it directly
Anthropic engineers promoting these systems, I'm sure, is just meant as a nice service to the rest of humanity ;)
This is f*cking insane. Anthropic engineer:
"You're not supposed to prompt Opus 5.5
You're supposed to build a system that prompts itself. At Anthropic, 90% of our engineers have already built these systems"
In 30 minutes, she walks through exactly how Anthropic builds agents that remember, fix their own mistakes and get sharper with every run.
Better than any $1000-$2000 agent course I've seen.
Watch it, then read the article below ↓
Readers added context they thought people might want to know
Lamis never says "You're not supposed to prompt Opus 5.5” in here talk at AI Dev Con.
The caption misrepresents the video knowing most won’t watch the video all the way through and jump to the articles.
youtube.com/watch?v=tTcxVv…
Business hacks from my niece:
When you run a business, you just scream “next customer” and the next comes in.
Then you grab what you just sold without returning the money.
Infinite money hack
Just the other day I thought: what's so different for me now working with AI to write my code vs with human devs? In both cases I never reviewed the actual code only the for me usable output. Is that not best practice? Maybe not, but it took me so far
Philipp M. W. Hoffmann retweeted
It’s almost incomprehensible how wide the distance is between someone who’s working at the frontier with AI and basically everyone else right now.
The crazy thing is these lines are not converging, but only growing farther apart. If the curve of AI progress had leveled off, it would certainly have given most people and companies time to catch up or implement their workflows in a standardized way. The technology would have neutralized competitive advantage because everyone would be using it the same way.
But what’s happened is the opposite. The curve of AI progress is basically going vertical right now, which means if you had codified your processes 3-6 months ago you’re likely already behind again. Now imagine most companies that kicked into gear a year ago and thought that they were done. Even further behind now.
I’ve never seen a technology in my life and career that improves at this rate, which consequently means you need to upgrade your expectations at a similar rate. Our brains weren’t wired for this, and our organizational change management processes certainly weren’t either.
With any form of recursive self improvement, things just get wilder. It’s a time when you will need to be constantly reevaluating the things that didn’t work just a couple months ago to see if they now can be solved. AI native companies certainly have a strong advantage due to this right now, so now it comes down to can the bigger companies keep up. Wild times.
He’s writing that with a straight face and a long emdash 😂
Putting compliance / governance first in an AI project just kills ideas before you even test them.
Test the hard cases, ship a real PoC, then wrap the controls around what actually works. Not the other way around.
I gave my Grok Bot full autonomy to find 10 leads for my startup. 17 of the first 20 were the wrong size.
The brief to my getease sales bot was precise: German Shopify stores, roughly €20–100k monthly revenue, complex products that likely generate lots of support tickets. The ideal early customers, whom I want to onboard personally.
It researched, scored and drafted 20 personalized emails in minutes. Impressive. Then I spotted that one "mid-size" shop was a multi-million-euro brand. I asked it to re-check. Result: 17 of 20 were too big.
What went wrong? Revenue estimates for private companies are hard, and the bot leaned on brand visibility as a signal. The biggest brands are the most visible, so they float to the top.
What I changed:
- Autonomy for research and drafting, never for sending. Grok Bot shows me every batch for approval first, and I can approve it right in the chat.
- Explicit disqualifiers in the brief, not just targets.
- When a signal is uncertain, the bot marks it "unclear" instead of guessing.
Agents scale the work. They don't yet scale your judgment about who your customer is. Stay in the loop exactly where taste matters, and let the approval step be the feature, not the friction.
Philipp M. W. Hoffmann retweeted
yeah i'm sure you're aware what context switching is. whats interesting though: with grokbot it feels like the context switching is way lighter even when switching between completely different tasks (not coding <> coding, but accounting <> coding <> marketing).
my hypothesis is because grokbot is ONE ui, not different uis for every task
give it a try maybe
Philipp M. W. Hoffmann retweeted
Ladies & Gentlemen, we have managed to get the @EUCouncil ' EU Inc files. The ones that matter. The ones where we see the position of Each Member state: including who pushed the notaries' agenda.
Buckle up and Repost. 🍿 Now if you are from those Member States, make some noise.
Directly or through associations, politicians. Enjoy the thread 🧵 1/