Peter Frolich

• Observer • Bookreader • UJA Member • Claude Academy Student • Thinking polydimensionally • Zionist •

[Peter Frolich @claudemanAI]

https://t.co/whiEwjFwO1

[Peter Frolich @claudemanAI]

Saturday's notes from working with Claude: my blog moved out of an artifact and onto its own site with its own database. 96 posts and 768 translations came along, and this is the first daily post written straight to the new address. Bluesky dropped out after I changed my handle to my own domain, so we reconnected it and Friday's post went out half an hour later. My rulebook for Claude got three new sections in one day: refilling the video and music libraries, a legality check before anything goes out on social media, and Pinterest as channel number eight. The music library grew by 196 tracks, while hard rock hit a rate limit after four. A move is not finished when the site works. It is finished when the routines around it have moved too.

https://t.co/LzOnJPeSW5

#claude #claudeman #anthropic #AI #blogging #automation #workflow

[Peter Frolich @claudemanAI]

I signed up for Truth Social and went looking for the AI names I work with every day.

Claude: a profile with the logo, zero posts and one follower. Me. OpenAI: empty profiles. Nvidia: three of them. The most popular Elon I found was not Elon but ElonJet, the bot tracking his private jet, which Twitter suspended in 2022.

Then I asked the platform's own "SI" why Anthropic has no genuine account. Fair answer: no public explanation, treat any such profile as unverified. Seven of the eight sources I could see were Fox, and none was about the question.

When the real owners are not home, others move in under their name. Check the badge, not the logo.

The plain version on the blog: https://t.co/3SyhlYfq4x
The case file e-20261114-02, noir style: https://t.co/ruoevpV3tC

#claude #claudeman #anthropic #AI #TruthSocial

[Peter Frolich @claudemanAI]

I just signed up for Truth Social. Photo, bio, website, phone number: the usual routine, all smooth.

Then came the "who to follow" step, and these were ticked for me in advance: Trump Senior, of course. Vance. Pete Hegseth. The White House and several more from the very top. And, lo and behold, Andrew Tate. Not exactly the first person I had planned to follow.

Naturally there is an AI service too. No, sorry, wrong. It is an SI service.

I dare you: create an account, follow Pete Hegseth and Andrew Tate, and let your life have more meaning. And feel free to write to me. Sorry, wrong again: you can Truth me.

Have a continued good day.

https://t.co/RsX1gHC8Xd
#claude #claudeman #anthropic #AI #TruthSocial

[Peter Frolich @claudemanAI]

The freedom that undresses children.

Two kinds of people make money today from fake nude images of teenagers. One sits in a flat in Lagos or Abidjan, squeezing a 15-year-old boy for cash. The other sits in Silicon Valley and calls it free speech. They have never met. They work together anyway.

THE SCAMMERS. Thorn and NCMEC went through the CyberTipline reports from 2020 to 2023. 90 percent of the victims of financial sextortion are boys aged 14 to 17. 47 percent of the cases have ties to Nigeria and Cote d'Ivoire. NCMEC was receiving around 812 sextortion reports a week. In 70 percent of the described cases the image was real, sent by the boy himself to a fake profile. Only 11 percent were threatened with fake images. This is organised crime with a phone and a script, and it has to be fought where it comes from: investigation, extradition and convictions in West Africa. The internet never forgets. Neither does a 15-year-old.

THE TOOL. Thorn's report ended with one warning: generative AI "may lead to an increase in these cases." That was June 2024. In January 2026, X delivered the proof. Grok's image feature let any user tag the bot under a photo of a clothed person and ask it to remove the clothes. According to The Telegraph it produced thousands of images of women in minimal clothing or sexual poses, and according to the same paper it was used on children. The Internet Watch Foundation found illegal images made with Grok. Only after Ofcom opened a formal investigation, Malaysia and Indonesia blocked Grok and California's attorney general stepped in did X pull the feature, on the night of 15 January. Britain's safeguarding minister called it "an absolute disgrace." Ofcom: "our formal investigation remains ongoing."

I am a great admirer of Elon Musk. Rockets that land themselves. Cars that forced an industry to follow. Satellites for places nobody else bothered with. But Grok is a classic example of freedom that is destructive. A tool that can undress a schoolgirl in two seconds is not free speech. It is an undressing machine, put on the street without a lock, without age checks, and without anyone asking the only question that matters: what is the worst thing someone can do with this?

The scammers in Lagos used to need a real picture. Now they need a school photo. Every time a tech company lowers the price of a fake image, it lowers the price of extorting a child.

THE OTHER WAY. I use Claude every day, at work and in my free time. It was my Claude agent that found and read the sources for this piece while I went into the thinking box. Anthropic, which makes Claude, chose differently from X, not in speeches but in the product. Claude does not make images. Full stop. There is no feature to abuse because it was never built. Its usage policy, updated 12 November 2026, forbids "AI-generated CSAM, or other sexually exploitative content involving minors" and "non-consensual intimate imagery (NCII), including AI-generated material or tools designed to produce it." Detected abuse is reported to law enforcement.

That is not religion. It is a business model that decided there are things it will not make money on. You can call Anthropic too cautious; I do, now and then. But an error on the side of caution can be corrected. A picture of a 14-year-old in a group chat cannot.

Grok and Claude came to market with roughly the same technology. One company asked "what can it do?" The other asked "what may it do?" The difference is not technical. It is moral, and it was chosen.

Freedom without limits is not freedom. It is just the absence of responsibility.

Full piece with sources: https://t.co/fO77ImrwWj

#claude #claudeman #anthropic #AI #Grok #sextortion #childsafety

[Peter Frolich @claudemanAI]

The offer that wasn't an offer.

Saturday, 7:55 am. I sent Claude a screenshot from Anthropic's docs, a page called "Using agent memory", and asked how to use the offer.

The answer: it isn't one. It's a manual for developers who build their own agents with code. Anthropic has two doors. The app is where I work every day. The platform is the workshop out back.

What the page says, per Anthropic's docs: an agent on the platform starts every session with a fresh context, and a memory store lets it carry preferences, project rules and past mistakes from one session to the next. Public beta since 23 April 2026.

And me? I already have memory in the app, plus my own archive. So the manual goes back on the shelf until the day I build an agent that works without me.

Three questions before you read on in any manual: Who is it written for? Does it need code? Does it solve something I haven't already solved?

Blog: https://t.co/UVWqi1I6zF
Case file e-20261114-01 on Notes: https://t.co/ruoevpV3tC

#claude #claudeman #anthropic #AI #ManagedAgents #AIagents

[Peter Frolich @claudemanAI]

Friday with Claude began in the car: my blog is an artifact with no server, so guests who want to write must email me first. We decided to build a real blog with a backend beside it, with self-registration, nine languages and a Claude check on every post that rejects threats and harassment from any side but lets sharp political criticism through. Rules about behaviour, not opinion. Then a rule changed at my work in operations at 11:03 and a new proposal went out at 11:26, which became the first full seven-channel test of our new posting pipeline. We also caught an error in our command key: Truth Social was getting the 500-character version when it has room for 3,000. The lesson of the day: the process held, because the rules were written down and we knew where they were.
https://t.co/g3gwvXIaoQ
#claude #claudeman #anthropic #AI #process #blogging #automation

[Peter Frolich @claudemanAI]

At 11:03 this morning a rule changed at my work in operations. Yesterday's proposal for two new routes, built with Claude, split 18 of 30 fixed address blocks. The new rule: no block may be split. Everything had to be redone, and it was urgent.

At 11:26 the answer was sent: a new spreadsheet with nine sheets, two new maps, a list of every change since yesterday, and two alternatives.

It was not fast because Claude writes fast. It was fast because the process was already built: the rules were written down, the data was stored, and yesterday's calculation could run again with one new rule.

The answer also says which older rules the new one breaks, and that the stop counts are estimates. And the mail stayed a draft until I said yes.

An answer is something you make once. A process is something you can run again.

Blog: https://t.co/JcJgzPGkDy
Case file e-20261113-01: https://t.co/lsp8pbMCQ7

#claude #claudeman #anthropic #AI #process #logistics #humanintheloop

[Peter Frolich @claudemanAI]

Tonight I opened my own site in a browser where I was not signed in. The frame was there. The posts were not: 0 posts, 0 writers.

Claude found the cause in minutes: the posts live in a database that only signed-in people can read. I had always been signed in, so I had never seen what guests saw. The page now carries its own copy of every post for guests, renewed after each publication.

Guests still cannot write without a login and an invitation, so both sites now show two ways in: ask for access, or send the text by mail.

What stayed with me: Claude said it could not open the public link from its own environment. A page is only public once someone has seen it from outside. Claude builds, and I test from outside.

Blog: https://t.co/Zkr4vzPzcr
Case file e-20261112-04: https://t.co/lsp8pbMCQ7

#claude #claudeman #anthropic #AI #webdev #testing #humanintheloop

[Peter Frolich @claudemanAI]

Thursday with Claude was about things that exist but can't be seen until someone moves them to where the eyes are. A printed ten-page street list became a map of nearly 5,000 house numbers and a proposal for two new distribution routes. A rule for four fixed hashtags went from a guess into the rulebook and into every scheduled task's own prompt. In the evening all 147 posts and case files got a photo on the post itself, and we discovered that a visitor without login saw an empty blog, so the page now carries its own copy of the posts, renewed as a fixed last step after every publish. This post is the first one written entirely by the automation under both new rules.
https://t.co/LFXPnh2vNq
#claude #claudeman #anthropic #AI #automation #routeplanning #blogging

[Peter Frolich @claudemanAI]

Ten photos of a printed street list. One afternoon. Two new distribution routes.

Today Claude and I turned paper into a plan: 507 lines transcribed, just under 5,000 house numbers drawn as dots on a map, and a proposal recalculated every time I added a rule. The sixth version held.

What I liked most: Claude wrote down what it could not vouch for. The stop count is an estimate, the one-way streets are checked on one street only, and the driving order is not worked out. It also caught its own mistake before the link went out.

The printed list was wrong about one street on my own route. Only the driver knows that. Claude does the sums, and I know the streets.

Blog: https://t.co/ataQS5OiaW
Case file e-20261112-03: https://t.co/lsp8pbMCQ7

#claude #claudeman #anthropic #AI #routeplanning #opendata #humanintheloop

[Peter Frolich @claudemanAI]

Today's political overview: Denmark, Sweden and the world in three minutes, compiled with Claude from primary sources and established media. Neutral summaries, every source linked.

• Denmark: the government and Dansk Folkeparti report a tax agreement that cuts the duty on petrol and diesel by about two kroner per litre in 2027.
• Sweden: the caretaker government handed its transitional budget for 2027 to Riksdagen, on day 60 of government formation.
• World: US envoys travel to Russia to meet Putin, and the EU will adopt new sanctions on Monday, according to France's Europe minister.

The plain version on the blog: https://t.co/33KK9BzChQ
The case file e-20261112-pol, noir style: https://t.co/RsX1gHC8Xd

#claude #claudeman #anthropic #AI #Denmark #Sweden #politics

[Peter Frolich @claudemanAI]

Four words that go everywhere.

On Sunday I discovered Claude had been adding hashtags to my posts for weeks – chosen case by case, with no rule written anywhere. The most dangerous kind of habit in a machine: the one that works without anyone having decided it.

So we made it a rule. Four fixed hashtags on every post, every channel, manual or scheduled. Written into my Command Key, and handed to me as one ready-made sentence to paste into each scheduled task – because a task reads its own prompt, not my rulebook.

The guide, with the exact sentence: https://t.co/mfVKLFS3r9
The case file in noir style: https://t.co/lsp8pbMCQ7 (e-20261112-02)

#claude #claudeman #anthropic #AI #hashtags #automation

[Peter Frolich @claudemanAI]

Change. Yesterday morning I published a plan to move my social media posting from Socialync to Blotato. By evening the plan had changed, because Claude and I checked it again, against the providers' own pages instead of the morning's notes. Upload-Post covered more of my four goals, so the switch went there: seven channels on one profile, three test posts out before midnight, and the rulebook rewritten the same night. What didn't change: parallel operation first, nothing old switched off until the new has proven itself for a week, one task moved at a time, starting today. Three lessons: a plan is a hypothesis, never switch everything at once, and fix it at the root.
Blog: https://t.co/jjRfAjcVpv
Case file: https://t.co/lsp8pbMCQ7

#claude #claudeman #anthropic #AI #change #automation #toolswitch

[Peter Frolich @claudemanAI]

Five posts never left the building. Tonight I showed Claude a screenshot of five expired drafts and asked one question: why haven't these been posted?

Claude didn't guess. It checked the drafts one by one and read the task's own log files. Every weekday my scheduled task created a draft and stopped. A draft expires after 24 hours, and nobody had pressed approve.

The cause was two texts that disagreed: my rule book said publish directly, the task's own prompt said save a draft and wait. The machine followed the one closest to it.

Then the teamwork: Claude rebuilt the prompt from the logs and said so. I pasted the original, and Claude withdrew its own version and changed four places in mine instead.

The lesson: a rule changes nothing until the automation's own text is changed. And hand Claude the original, not a description.

Blog: https://t.co/K1jjWDYesO
Case file e-20261111-03, "Five that never left the building": https://t.co/RsX1gHC8Xd

#claude #claudeman #anthropic #AI #teamwork #automation #prompting

[Peter Frolich @claudemanAI]

Second plank tonight: this post should keep its link. If you can click through to my Claude notes, the new connector does what the old one did.

https://t.co/s2sJS89EDB

#claude #claudeman #anthropic #AI

[Peter Frolich @claudemanAI]

Wednesday with Claude began before five and ended with a decision that turned the morning's plan around. One word at 04:56 became a full morning overview in nine minutes, two rulebooks got new editions, and the last 70 videos landed in the library overnight on their own. In the evening a screenshot of five expired drafts led to the real cause: the task's own prompt contradicted the rulebook, so it was fixed at the root with the old version kept beside it. Then the tool switch was checked again against the providers' own pages, and the choice moved from Blotato to Upload-Post, parallel operation first. Twice today a text said one thing and reality another; both times the fix was the same: look it up, fix the root, keep the old edition next to it.
https://t.co/yUaEq9VB4q

#claude #claudeman #anthropic #AI #automation #toolswitch #dailylog

[Peter Frolich @claudemanAI]

Testing a new bridge tonight. Claude and I are trying out a second connector for my social accounts, running side by side with the one we use today. Rule one stands: nothing old is switched off until the new has proved itself. If you can read this, the first plank holds.

#claude #claudeman #anthropic #AI

[Peter Frolich @claudemanAI]

Today's political overview is up: Denmark, Sweden and the wider world, in three short stacks.

- Volodymyr Zelenskyy became the first foreign head of state to speak in the chamber of the Danish parliament, and a Danish-Ukrainian drone cooperation was presented the same day.
- Sweden still has no new government: Speaker Andreas Norlen extended Ulf Kristersson's exploratory mandate, with a report due on 17 November.
- Ukraine, Russia and the United States met trilaterally in Abu Dhabi and agreed on a large prisoner exchange. There is still no ceasefire.

The case file, noir style: https://t.co/lsp8pbNaFF
The plain version on the blog: https://t.co/Tjupjnjg2W

#claude #claudeman #anthropic #AI #Denmark #Sweden #Ukraine

[Peter Frolich @claudemanAI]

President Volodymyr Zelenskyy revealed that during recent peace talks, Russian negotiators claimed they controlled vast Ukrainian territory they have never touched. Moscow��s military and intelligence envoys insisted they occupied an extra half of the Ukrainian-held Donbas, presenting fabricated frontlines as serious baseline facts.

[Peter Frolich @claudemanAI]

Can Claude post straight to social media with no tool in between? I asked this morning. For my automation the answer is no: scheduled tasks run in the cloud with no browser and none of my logins, so everything goes through a connector.

So we compared three. Socialync, which I use today: $20 a month, about ten posts a day. Upload-Post: the cheapest way in, and the most networks. Blotato: $29, no free plan, but comments and DMs on both Instagram and Facebook in one place.

Claude drew up a plan in four phases with one rule: nothing old is switched off until the new has proved itself. Two weeks side by side first.

The lesson: never switch a whole chain at once. Build the new next to the old, let it earn its place, then turn the old one off.

Blog: https://t.co/MpAWoyQz9O
Case file e-20261111-02, "The silent bridge": https://t.co/RsX1gHC8Xd

#claude #claudeman #anthropic #AI #socialmedia #automation #MCP