
Sign up to save your podcasts
Or


Fable 5.2, Nvidia’s investigation, and Claude’s office tools: Fable 5.2 is here—but how much of the excitement holds up in everyday work? Thomas Bekkers and Rudy Faile discuss the model, the investigation surrounding Nvidia’s Jensen Huang, and Claude’s growing role in the workplace.
Recorded November 11, 2026. The espionage claims discussed in this episode are allegations, not established findings. “Claude Office” is our shorthand for Claude’s workplace tools, not an official product name; predictions about future products are our opinions.
Hosts: Thomas Bekkers @thomasbekkers and Rudy Faile.
More from Thomas: https://www.youtube.com/@thomasbekkers
00:00 This week on AI Psychosis
01:00 Fable 5.2: real use versus the hype
12:30 Nvidia, Jensen Huang and the allegations
18:43 Claude as the workplace
25:20 Final thoughts
Host: Thomas Bekkers on YouTube
Watch on YouTube: https://youtu.be/tZ_hB6oqba4
Grok Bot usage limits, the Binance hack, and European freezer attacks: Grok Bot usage limits have tightened, Binance is responding to a major security breach, and supermarket refrigeration systems in Europe have been targeted. Thomas Bekkers and Rudy Faile discuss what the stories mean for people building with AI and relying on connected infrastructure.
They discuss the economics of AI agents, the Binance incident, the risks of exposing connected equipment to the internet, and the potential role of AI in cybersecurity.
Hosts: Thomas Bekkers (@thomasbekkers) and Rudy Faile.
Chapters:
00:00 Introduction
00:56 Grok Bot usage cuts and trust
06:06 The cost of AI and paying for more usage
10:14 Binance's $6.3 billion hack
15:00 Europe's freezer attacks
19:28 Can AI improve cybersecurity?
23:14 Closing thoughts
Sources discussed:
Elon Musk on Grok Bot serving costs: https://x.com/elonmusk/status/2116992258520064397
Binance's incident update: https://www.binance.bh/en/blog/community/3229153443759566763
French cybersecurity alert on refrigeration controllers: https://cert.cyber.gouv.fr/alerte/CERTFR-2026-ALE-012/
Claroty's research on refrigeration controller vulnerabilities: https://claroty.com/team82/research/freeze-the-controller-defrost-the-food-uncovering-vulnerabilities-in-danfoss-refrigeration-controllers
Host: Thomas Bekkers on YouTube
Watch on YouTube: https://youtu.be/J2szhOyuQfU
AI cyberattack prepping, Sam Altman on bugs AI can't patch, GitHub Octoverse AI coding agents and the ChatGPT medical-advice lawsuit: Thomas Bekkers and Rudy Faile talk about prepping for an AI-driven cyberattack, Sam Altman's warning that "we can't patch all the bugs," 40% of GitHub pull requests now starting with AI agents, and the ChatGPT medical-advice lawsuit.
Hosts: Thomas Bekkers (@thomasbekkers) and Rudy Faile.
0:00 Intro
0:50 Why we're prepping for a cyberattack
2:00 Sam Altman: "We can't patch all the bugs"
4:15 What to stock up on
6:46 40% of GitHub PRs are AI
9:29 Banks still run on decades-old code
11:41 How cache timing attacks work
14:15 Hardware vs. firmware bugs
15:07 ChatGPT told her to hide it from her doctor
20:03 Lawyers caught by fake AI citations
22:57 Final advice: stay prudent
WHAT TO STOCK
Water: 1 gallon per person per day (Ready.gov). The Red Cross suggests 2 weeks of supplies at home.
Food that keeps and needs no cooking, a week of medications, batteries, a power bank, a hand-crank or battery radio, and some cash.
This isn't a reason to panic. It's a one-time expense, so you have a backstop.
SOURCES
Ready.gov emergency kit: https://www.ready.gov/kit
GitHub Octoverse coverage: https://thenextweb.com/news/github-universe-copilot-fusion-octoverse-agents
ChatGPT lawsuit coverage: https://thenextweb.com/news/parents-sue-openai-chatgpt-seizure-medication
The lawsuit's claims are allegations that no court has ruled on yet.
Host: Thomas Bekkers on YouTube
Watch on YouTube: https://youtu.be/cTmFL3b9Ma8
PHP security vulnerabilities, eMudhra certificate trust, LiteLLM and AI vulnerability research: Thomas Bekkers and Rudy Faile discuss PHP security vulnerabilities, eMudhra certificate trust, LiteLLM, and how AI is changing vulnerability research.
Hosts: Thomas Bekkers (@thomasbekkers) and Rudy Faile.
0:00 PHP security discussion
1:24 How researchers disclose vulnerabilities
3:44 Hosting providers and mitigations
5:24 What website owners can do
8:30 eMudhra and certificate trust
11:53 LiteLLM and software dependencies
14:12 AI and vulnerability research
17:35 Closing thoughts
WHAT TO DO
If you run PHP yourself (VPS, dedicated server, or your own containers), upgrade now.
Patched versions: 8.5.12 / 8.4.27 / 8.3.36 / 8.2.35.
If you're on a managed host, check your host's status page or blog and confirm the update has been applied.
THE VULNERABILITIES
CVE-2026-109618: remote code execution via out-of-memory corruption.
CVE-2026-109619: remote code execution via stale global state.
Both are rated CVSS 9.2 (Critical).
SOURCES
PHP Foundation: https://thephp.foundation/blog/2026/10/21/todays-php-security-release-and-how-were-strengthening-php-security/
Automattic: https://automattic.com/2026/10/21/php2shell/
Host: Thomas Bekkers on YouTube
Watch on YouTube: https://youtu.be/fPvT1nY0JqU
Anthropic, SpaceX, universal basic income, AI jobs, coding and personal AI tools: Anthropic has announced a Public Benefit Fund, and Elon Musk has pledged SpaceX shares for children in America. Who benefits, who decides, and what would it mean for a future with fewer jobs?
In AI Psychosis Episode 4, Thomas Bekkers and Rudy Faile discuss the announcements, then debate deterministic code vs. neural networks, learning AI by building, and personal tools: camera autofocus and a faster smart TV experience.
The announcements are commitments, not completed distributions or a universal basic income program. Anthropic’s initial 5% share commitment is subject to stockholder approval. Musk’s pledge concerns one of his SpaceX shares per child in America in 2027; implementation details remain unclear.
Hosted by Thomas Bekkers and Rudy Faile.
Thomas: Thomas Bekkers on YouTube
CHAPTERS
00:00 Anthropic and SpaceX: sharing the wealth
02:18 Who benefits, and what are the risks?
07:48 Deterministic code vs. neural networks
14:43 Learn AI by building something
18:53 AI, brain rot and critical thinking
24:14 Building a camera autofocus tool
25:45 Fixing slow smart TV apps
26:47 Smart home security and network segmentation
28:03 Rebuilding the Samsung TV Plex experience
29:42 How skills carry over between projects
33:47 Rudy’s open-source TV app and wrap-up
ORIGINAL SOURCES
Anthropic’s Public Benefit Fund announcement (October 18, 2026):
https://www.anthropic.com/news/public-benefit-fund
Musk’s SpaceX share pledge (October 19, 2026):
https://x.com/elonmusk/status/2112089556615123114
AP reporting on the pledge:
https://apnews.com/article/b3312da37a0e340be275295073055e00
Watch on YouTube: https://youtu.be/tg5b9RYehGU
Muse Wonder vs Gemini 4 Argon, AI benchmarks, Omarchy, Spacefast and AI agents: Meta's Muse Wonder and Google's Gemini 4 Argon launched a day apart, both with big benchmark claims. Muse vs Argon: is it all hype?
In episode 3, Thomas Bekkers and Rudy Faile share first impressions of Meta's Muse Wonder for coding, talk about why self-reported benchmarks tell you little until you've used a model for a few days, and ask whether the frontier labs are converging and whether open-weight models can keep up. We get into faster tokens, usage limits, and letting agents work overnight. Then we look at Omarchy, DHH's Linux-based operating system, and the new Familiar plugin that makes it feel more like a Mac, plus why Linux suits AI agents. We close on AI-generated music, Spacefast (Automattic's new product for putting anything online straight from your agent), why we stop using software our agents can't use, and the faster Plex app Rudy built for his TV in 45 minutes by talking to an agent.
Benchmark results discussed are the companies' own claims. Rudy works at Automattic, which makes Spacefast.
AI PSYCHOSIS | Business Meets Tech
Conversations about what we're building, what's useful, and what's overhyped.
YOUR HOSTS
Thomas Bekkers (@thomasbekkers) — Serial Entrepreneur and AI Integrations Specialist.
Rudy Faile — Head of Security, Automattic.
CHAPTERS
00:00 Gemini 4 Argon and Muse Wonder drop a day apart
00:54 First impressions of Muse Wonder for coding
03:07 Can you trust self-reported benchmarks?
05:05 Codex, Opus 5.5 and the models we actually use
07:01 Faster tokens are the next frontier
08:14 Letting agents work while we sleep
09:32 Are the frontier models converging?
11:17 Can open-weight models keep up?
13:24 Omarchy goes Familiar: Linux that looks like a Mac
14:59 Publishing iOS apps without a Mac
15:57 How we use Omarchy every day
17:07 Why AI agents feel at home on Linux
19:33 AI-generated music is getting good
21:33 Spacefast: put anything online with your agent
26:14 No API or MCP? We stop using it
27:54 Rudy rebuilt his TV's Plex app in 45 minutes
29:44 We're still early: just start talking to agents
Watch on YouTube: https://youtu.be/nRGyor1nsKU
Claude Code, CLAUDE.md, AGENTS.md and AI agent setups: Are we overbuilding our AI agent setups?
In episode 2, Thomas Bekkers and Rudy Faile talk about why lean CLAUDE.md and AGENTS.md files and plain coworker-style prompting are replacing bloated agent setups, and what still belongs in skills. We compare Handy and Wispr Flow, ask whether AI reverse engineering is putting closed-source software and SaaS at risk, and discuss Elon Musk's announcement that Grokbot will route tasks to the best backend model, including Opus 5.5 (announced; the rollout is unverified). We also look at Kindlefall, the action RPG Rudy built by talking to an AI agent while he played it, and we close on P(doom) versus P(gloom).
AI PSYCHOSIS | Business Meets Tech
Conversations about what we're building, what's useful, and what's overhyped.
YOUR HOSTS
Thomas Bekkers (@thomasbekkers) — Serial Entrepreneur and AI Integrations Specialist.
Rudy Faile — Head of Security, Automattic.
CHAPTERS
00:00 Are we overbuilding our agent setups?
03:19 Talk to your agent like a coworker
04:58 What still belongs in AGENTS.md and skills
08:10 Handy vs. Wispr Flow
09:54 Is closed source software dead?
12:56 Open source the app, sell the cloud
14:32 Tiny teams, billion-dollar companies
15:33 Rewriting software in Rust with AI
16:53 Grokbot and Elon's model routing
20:11 AI agents in Slack at work
21:49 Rudy built a game by talking to AI
27:15 Can one person make a AAA game?
29:08 P(doom) or P(gloom)?
Watch on YouTube: https://youtu.be/53Dpji7xuQY
AI coding, junior developers and AI agents: AI can write the code. So what's the developer's job now?
In our first episode, Thomas Bekkers and Rudy Faile talk about building with AI agents, the future of junior developers, and whether a computer science degree still makes sense. We compare the tools we actually use, voice workflows, AI hype, and the trade-off between giving agents more access and keeping control.
AI PSYCHOSIS | Business Meets Tech
Conversations about what we're building, what's useful, and what's overhyped.
YOUR HOSTS
Thomas Bekkers (@thomasbekkers) — Serial Entrepreneur and AI Integrations Specialist.
Rudy Faile — Head of Security, Automattic.
CHAPTERS
00:00 Welcome to AI Psychosis
00:15 How we got hooked on building with AI
03:46 AI hype versus real use cases
06:27 AI agents for security research
10:33 From AI skepticism to daily use
15:28 Linux, open source, and customizing your computer
18:38 Faster AI and the human bottleneck
23:47 Voice workflows: stop typing everything
25:44 What happens when AI writes the code?
30:37 Junior developers, soft skills, and college
35:20 AI models versus agent tools
40:09 Running specialized agents in a team
47:44 Letting an AI assistant handle the inbox
53:51 Learning to judge AI output
57:21 Goal mode, planning, and keeping agents working
01:04:47 How much access should your AI have?
01:09:41 Staying healthy while building
01:11:35 Just start building
Watch on YouTube: https://youtu.be/csPfBcHsHik