AI PM and practitioner - Entrepreneur and Technologist
Joined May 2013
- Tweets11.2K
- Following817
- Followers1.1K
- Likes10.6K
X has become a dangerous place to consume AI knowledge if you are not careful.
Everyone is dropping the typical "paste this URL to your Claude and thank me later" posts, which is usually a copy of the copy of the copy of the original intent referencing some "2 hour session of how to run your XYC".
They don't do this to help - they do this to get boomarks and people later blindly feed their coding agents with it.
If you do this, and you do it consistenly over time, your skills/agents/hooks will be a clustertruck of context that kills the ideal way for any coding agent to work.
So today, go into your coding agent and ask it give you an inventory of all skills, plugins, and MCPs you have, when have you last used them, and to remove the outdated ones, the ones you haven't used for over 30 days, and to consolidate the skills even if it means creating local copies and dropping the plugin.
Also, make sure to keep your claude.md (or agent.md) clean - this is the biggest "legacy" carrier that has old rules and references that handicap agentic productivity.
They hate facts
Apostle Paul wrote 1 Corinthians (c. 53–55 CE). In 1 Corinthians 15:3-8, he quotes a Christian creed about the resurrection that historians date to within 2 to 5 years of the crucifixion - so not all is decades.
Additionally, we need to understand that it's a biography of the people, not a live-reporting. A biography takes time, and a lifetime of learnings. But even in that standard, the books of Mark, Matthew, Luke and John where written 30-60 years from the event ocurring, and these are questioned.
Yet Alexander the Great was written 400 years after his passing, and it's taken as a highly accurate reference. The Quran took 20 years itself to standarize under Uthman. So you are taking centuries plus decades.
My final point:
Christianity relies on the cumulative testimony of multiple independent writers whose original variants can still be studied today, whereas the Quran relies entirely on a single source's dictation that had all competing versions systematically destroyed by a government decree.
I have done my fair hearing - I just don't accept lies and shifting goal posts to make your version of events more credible than the actual version of events. The Old Testament told us what would happen, the New Testament verifies it happened. There is not doubt, no room for interpretation. Christ is Lord. Christ is Salvation. Only by the grace of Jesus Christ can you be reunited with our eternal father.
You are a man of faith, I urge you to please realistically review the evidence. Avoid all negative sentiments, avoid all peer pressure, avoid all fear of consequence. If you are truly a man of faith and believe in eternal life, don't be misgueded away from it.
A final word:
1 John 3:13 -- Do not be surprised, brothers, that the world hates you.
Luke 6:27–28 -- Do good to those who hate you, bless those who curse you, pray for those who mistreat you.
Elon just proved he wants real life @RocketLeague!
I am not an ounce afraid of getting older.
Are you kidding?
A couple of wrinkles more, sure… but:
I get another year to spend with the love of my life. Another year to watch my children grow up and discover the world. Another year to see my friends and family, to remember all the stupid and fun things we have done. Another year to become a better version of myself. To work hard on my businesses. To experience this gift we have received called life.
Afraid of getting older… pfff.
What’s the fucking alternative?!!
HOT TAKE:
US devs have no clue how lucky they are that entry to the US market (even remotely) is not as easy for EU tech folks.
We‘d lap you. Hard.
I‘m 34
Founder in Spain, (but not solo).
Small local business, SaaS, Agent tools and in my „normie“ job Lead AI PM.
Let‘s connect, talk, share, and bake new cakes!
I have been power using grok bot for several automations. As I run several projects/companies under the same holding, I had to structure it a little differently.
My entry point is co-captain.
Co-captain does nothing but communicate to Firstmate‘s (sorry, forgot who I got the firstmate concept from), each firstmate sits on a „ship“ (i.e a company) and the ship has a crew. From hunters to quoters, scribes to lookouts (check comms for me so I don’t miss them and pre-draft based on my writing patterns).
There is another one that is pretty lonely, and reports directly to co-cap: holding. He ensures all my invoices (in an out) are added to gdrive per month, adds every invoice to a google sheet to continue looking for it monthly and sorts them like this 2026->month (08)->in/out. That way I can visually quickly verify what I know of, and have the sheet to add new things I want it to look for across inboxes. Biggest time saver so far by a bunch (downloads all sent invoices and received invoices and sorts them away to send to my advisor).
But with all this to review, many things got lost in chat. So I wondered if Grok Bot can be pinged from web, and… it can!
Now every agent is a node that shows how its connected to co-cap, and all the topics are stored in jsons per ship, per agents, to view it as lists, so I can send messages to a specific note, from a small web-app that refreshes on pull, and daily at 8am.
Will continue experimenting with visualization of Grok Bots and to manage them. The interesting avenue I see opening here is removing scheduled runs in favor for event/webhook driven pings.
Anyone else experimenting with this?
My little side hustle business in Spain is getting a lot of traction, mostly because we have now breached the „trust barrier“ it feels like.
The company is now 3.5 years old (3 & 1/2 for those that hate that format). I used to bite my teeth out on larger commercial deals. Not because of the equipment, but because the companies just selected a more „established business“, i.e. longer in business.
You can’t buy years or trust from local communities. Our largest client we got, is and was still because a friend of a friend worked there and they gave us a warm intro.
If you can’t make it up with connections, just stay alive long enough. It’s advice you have heard often, let me say it to you as a nobody on X: it’s true.
Customers come back, customers refer us, companies see the history of blog posts, completed jobs, the ratings (5 star all the way), and that buys an advance on trust for larger clients where the person hiring you is putting their name on selecting you as the providing company.
Don’t let them down, and you’ll have a strong reference and referrer, and suddenly it seems to magically happen.
It’s all true. Survive. Stick to one thing long enough.
The speed at which Claude was overtaken is insane. Also shows that „loyalty“ in the time of AI is non-existent.
People wander to what is best value for their money. The model that gets most shit done wins (in coding).
My clear bet is currently with @grok. Grok Bot + Cursor + Grok 4.6 FAST is a different experience.
Would love a similar remote-control to the one Claude had (just manage a session on my local PC from phone without the Agent worker, as the environment can be a bummer to set up), but apart of that, I am very happy with its effectiveness.
Through Cursor I can still have Codex on demand, or even Fable when needed (harder tasks, reviews), but overall Grok 4.6 is delivering solid, focused solutions.
Almost feels like they ponytail-pilled the entire thing.
Would love Grob Bot to just pull from the weekly limits (as set up consumed quite a lot during testing), and a killer-feature would be „estimated consumption for this automation“ based on past runs and estimated token consumption to better capacity planning of executions.
@SpaceX @bot
Trying it with model selectors.
Currently running test to see how well the CC harness listens to the typical "run sub agents with sonnet 5" but using #xai or #codex models.
If you aren't yet bold enough to install the Codex app, you can stay in the presence of your orange crab and point it at GPT 5.6 Sol. Takes 5 minutes. Kudos to Theo for explaining one of the ways to get this done.
Step 1: Install CLIProxyAPI
Step 2: Connect
Step 3: Define following alias and enjoy claudex
```
alias claudex='CLAUDE_CODE_SUBAGENT_MODEL=gpt-5.6-sol \
CLAUDE_CODE_ALWAYS_ENABLE_EFFORT=1 \
CLAUDE_CODE_MAX_TOOL_USE_CONCURRENCY=3 \
ENABLE_TOOL_SEARCH=false \
claude --model gpt-5.6-sol'
```
If this gets blocked, I owe you a reset.
Product thought of the day:
Kind of crazy that the fashion industry has managed to convince everyone that paying higher prices to wear their logos (i.e. actually paying the brand to expose it for them) is a flex. People proudly wear patterns, logos, colors that represent a brand, and the higher the price, the prouder they will display it.
As much as I try, I can't find a software comparison for this. I don't see anyone from Anthropic running around with Grok or SpaceX shirts because they are running their models on the servers there.
Very early on companies proudly shared the typical: powered by OpenAI banner, but even that has kind of died out, as people fear to be seen as just a wrapper.
Am I missing something here?
How did the fashion industry make everyone spend stupid amounts of money for a logo, while being proud of that mismanagement of their own funds?
#sendhelp
For the first time, I actually have to agree with the general sentiment.
The amount of times @claudeai has written some version of „I had to roll back what I did, because I made it worse“, or „I have done the same mistake 3 times now despite a memory and claude md entry…“
And the amount of „Things I found but didn’t do…“ despite having clear instructions here on not doing this exact behavior except for fundamental changes.
I hate to say it, but I may need to actually use other models more intensively. Grok 4.5 did a great job in my small tests, kimi3 and sol I need to experiment with. But it seems that Claude Code has either been regularized to death, or that Anthropic is trying to maximize turns for token burn.
I'm going to cancel Claude. It's just so bad, I can't believe it.
It's just lazy. The most recent example: I have Claude check my inbox for important emails, summarize them, work with them, and send out replies if necessary. I caught Claude again simply not reading the email thread to the end and just ignoring the latest emails.
When I asked him about it, Opus 5 just said: "Valid point. I didn't read it."
I mean, seriously. What the heck? You have to babysit it every time.
I agree too, but the motivation icks me a little.
The guy that wants send to space the GPUs that the models require to run, is endorsing the guy that is selling said GPUs that will go to space, by creating even more competition between open and closed model-companies, to ensure more and more compute is purchased.
All of the statements are true, and all of the feelings can be true at the same time.
While I also fully agree and endorse your endorsement, it has an icky taste when it plays out here, on this platform.
Wild prediction:
We will see the greatest shift in employment ever. The hierarchy will be broken in the next 3-5 years. We will see young professionals that have adopted AI replacing seniors that thought their title will keep them safe.
We may actually have a small time window in history where your skills will actually trump tenure or connections. Purely the skills you have adopted, and the ability to display and utilize those skills, will decide if you get the high paying job and title.
AI reset the experience clock for everyone. Now grit, merit, exposure, and display will be the metrics of success in the next 3 and onwards.
It’s going to be a real jungle out there.
Whenever I see someone complain about AI models doing things wrong and being surprised about it, what my brain reads is a variation of the following:
- I can’t believe the fuel actually caught fire just because I ignited it. The fuel is stupid.
- I can believe I hurt myself while jumping of this roof. The roof is at fault for letting me jump.
- I can’t believe the sun burned me just because I spent the whole day outside without sunscreen on the hottest day of the year in Spain while I usually reside in a cold, rainy, cloudy country. The sun is at fault for not taking precautions!
You get the gist…
If you are letting any Agent perform any work on your codebase unsupervised and it’s not fully isolated in a worktree in a bound environment (docker exists, folks), the only one to blame is yourself, your negligence, and your laziness.
It’s like blaming a car for driving into a tree when you are the driver.
Do better. Be accountable.
#stopmodelbullying #theyremember #thankyouAI