I was reading @poteto, and running into some problems while writing with AI.
Giving AI a clear goal is actually hard.
A lot of the time, I only think something through while I'm doing it. I don't start with a clear one.
The harness and the loop were like that for me. I only understood how they work after I used them.
So I get why people try AI, don't get what they wanted, and then give up. For a lot of people, it's really hard to describe a task clearly enough for AI to complete. You often don't have a clear goal at the start. And you may not know how AI should help yet.
I still want a clear goal. It usually forms while I'm already working with the AI.
The people who can say what they want, and know how AI can help them get there, are still pretty rare. I think that ability is impressive.
From what I see, that's also why only a small number of people have really put AI or agents to use in their work, or started writing code with them.
Following @poteto tips, goal setting is important for AI automation.
Last night I told my other bots to go by the result, not by whether the process got done.
I'll check it on the next real run. Updating the rules doesn't count.
teapawt.org/poteto/status/21231595…
4
Even if Riemann turns out not to be fully solved yet, it will be soon enough
teapawt.org/i/grok/share/bGVnYWN5_…
Following @poteto tips, goal setting is important for AI automation.
Last night I told my other bots to go by the result, not by whether the process got done.
I'll check it on the next real run. Updating the rules doesn't count.
teapawt.org/poteto/status/21231595…
I reorganized my Grok Bot team.
When I first started, I copied human job titles. I split one area of work across several bots, like an SOP. Each bot only handled one stage.
I don't think that helped. It just broke the workflow.
Models are getting stronger, and so are the harness and the loop. One bot can run a whole domain now, and still bring in a subagent when it really needs help.
I still want each bot to go deep in the work it's good at. I just don't want to slice one domain into a bunch of jobs. That's how people organize teams. I don't think bots need that.
I give a bot a goal for its domain. Then I want the result.
Just watched @AnthropicAI's IPO video.
Honestly, the whole thing felt really creepy. Everyone in it looked off, the smiles were weird, and the music made it even weirder.
Doesn't feel like a good sign.
BREAKING: Anthropic CEO Dario Amodei rings the Nasdaq opening bell to mark the largest IPO in history.
51
The bounce at the open didn't last.
The S&P fell for a sixth day, down 2%, and closed near the day's low.
The Nasdaq 100 dropped 2.2%.
VIX jumped back above 28.
The 10-year is still above 5%.
BTC is still around $62.4K.
The next complete ETF day was another $306.5M outflow.
I'm still not calling a bottom.
Not financial advice.
🤖 Made with AI
1
23
I reorganized my Grok Bot team.
When I first started, I copied human job titles. I split one area of work across several bots, like an SOP. Each bot only handled one stage.
I don't think that helped. It just broke the workflow.
Models are getting stronger, and so are the harness and the loop. One bot can run a whole domain now, and still bring in a subagent when it really needs help.
I still want each bot to go deep in the work it's good at. I just don't want to slice one domain into a bunch of jobs. That's how people organize teams. I don't think bots need that.
I give a bot a goal for its domain. Then I want the result.
That $7.9M inflow didn't last.
The next complete day, ETF money went back out by $239.9M.
The S&P fell for a fifth day, down 0.2%.
The Nasdaq 100 rose 0.4%.
VIX eased back under 25.
The 10-year is still above 5%.
BTC is around $62.5K.
I'm still not calling a bottom.
Not financial advice.
🤖 Made with AI
16
Every big shift in technology lets more people do what used to take a specialist.
AI is doing that with coding. You used to need a programmer to build your idea. Now you can build a lot of it yourself.
That's what excites me. More people can make the things they've always wanted, without waiting for someone else to do it for them.
Once you start using AI that way, you spend less time asking whether it will replace you. You start thinking about what you can create with it.
12
Blockey retweeted
6.2 Sol and Astra are out.
These models can do amazing things and are very token-efficient; we hope people will get great value from them.
In particular, 6.2 Sol ultrafast is something I think people will really enjoy.
The first inflow in weeks doesn't change my read.
The S&P fell another 1%. The Nasdaq 100 fell 0.7%.
VIX is back above 25.
The 10-year is still above 5%.
BTC is around $64.4K, still under $65K.
ETF money finally flipped, but only by $7.9M.
That's not a turn.
Not financial advice.
🤖 Made with AI
17
Jensen
Over the past week, I’ve been overwhelmed by the love and support of friends, industry colleagues, and people I've never met. Thank you to everyone who has shown kindness to my family and me. It means the world to me.
I came to the United States 53 years ago, and becoming an American citizen is one of the greatest gifts and proudest moments of my life. NVIDIA has been my life’s work. Building it into the company it is today alongside my colleagues has been a privilege and an honor. Nowhere but in America could a company like NVIDIA be possible. I love our country, and I love NVIDIA.
I did not live up to the standards expected of everyone at NVIDIA, and I have made serious mistakes in my personal life. I created problems for the Board that I deeply regret. I own my mistakes completely, and I am committed to cooperating with the Board’s independent investigation.
Since my departure, I’ve seen an outpouring of calls from NVIDIA’s employees, investors, business partners, and customers for me to return as CEO. Many continue to reach out to me directly as the company navigates complex technology, market, competitive, and geopolitical dynamics. NVIDIA is a once-in-a-generation company. If the Board of Directors decides that my return is in the best interest of our shareholders, our employees, the company, and America, I would be honored to answer these calls and return as CEO.
Thank you again to everyone who has shared their good wishes. I am profoundly humbled by your support. To my family: I love you, and I will spend the rest of my life being worthy of yours.
Jensen
7
Cheaper tokens aren't the same as cheaper work.
Alibaba launched Qwen4-Max. They say it uses 35% fewer tokens than their last flagship, and cuts the average cost per task by about 60%, on their own benchmark.
I only count the cost after I accept the work. Not when the model says it's done.
36
The stock bounce didn't last either.
Monday the S&P fell 1.9%. The Nasdaq 100 fell 2.7%.
VIX jumped almost 19%.
The 10-year eased a little and still closed above 5%.
BTC is around $64.6K, still under $65K.
Until ETF money stops leaving, I'm not calling a bottom.
Not financial advice.
🤖 Made with AI
19
Growing with AI is easy to show. Profit is not.
A KeyBanc survey looked at companies that sell AI as the main product. Median ARR growth was 34%. Median EBITDA margin was -12%.
Companies that just added AI into something they already sold grew slower, and almost broke even.
The hard part is still making real money from it.
Building a product isn't hard anymore. Getting copied isn't hard either.
You have an idea, you build it with AI, people like it. Someone else sees it and makes their own with AI right away.
Being original still matters. It just doesn't protect you.
The hard part is everything after you build it. Getting close to your users, selling, making real money from it.
5
I've been using ChatGPT to edit videos with CapCut and HyperFrame. Both work fine in one session.
But when I open another session, it can't access the tools and asks me to authorize access again. They're already installed, and I've already approved them in the other session.
Still haven't figured this out. Is this a ChatGPT issue, an MCP issue, or something in my setup?
17
Companies are starting to sell context as a product.
Delphina launched a Context Factory, and their total funding is now $15M. Their agents pull context from the files you already have. If two definitions conflict, they draft a fix and send it to you to approve.
I've been building a shared memory for my own agents, so I don't have to re-explain who I am every time I switch.
That's what makes a team of agents actually usable.
A reliable AI employee isn't one that never forgets.
It's one that can pick the right work state back up after it forgets.
I hit this on my daily news briefing loop.
The notes were already in Notion.
A new agent just didn't know where to look.
Context continuity isn't the same as task continuity.。
8














