Onoufrios Malikkides

A lot of devs want to get into AI but don’t know where to start. The truth is you’re closer than you realize. I broke down what actually matters and how to level up without the guesswork. onoufrios.substack.com/p/ec1a3cff-960…

@noahzweben Can I run an existing skill/command?

You can now schedule recurring cloud-based tasks on Claude Code. Set a repo (or repos), a schedule, and a prompt. Claude runs it via cloud infra on your schedule, so you don’t need to keep Claude Code running on your local machine.

- 2025: @github is down I'm going to take the day off - 2026: #ClaudeCode is down I'm going to take the day off

@gumsays @grok fact check

if you're debating a move to portugal remember: pros: - 0% tax on longterm crypto gains (held >365 days) - sunniest country in europe - ocean is always a few minutes away - no missiles flying over you, ever - amazing food cons: - very expensive housing and rents - expensive groceries, restaurants getting there - bad, slow public services - horrible bureaucracy for businesses overall it's a 10/10 if you are financially stable

2 of the 3 AWS availability zones in the UAE were knocked down! I think this is the main justification to put data centres in space. They are too critical and too easy to target on earth @elonmusk @SpaceX

@grinich Marginal cost of writing code is going to zero but cost of writing software still remains high otherwise your clients could just write code to build scim and sso

A new class divide is forming in engineering. And it's measured in tokens. $20/mo for basics. $200/mo devs are delegating to agents. $1,000/day AI is the dev. It's called the Jevons Paradox, and it's already playing out. Article in reply👇 #codex #ClaudeCode #jevonsparadox

Key contradiction from this week's pods: Is SaaS dead or just evolving? "SaaS is dead. Systems of record will die in an agentic world. AI destroys data switching costs next." — Siworski on @20vcFund vs. "The SaaSpocalypse is selective. Complex platforms with accumulated edge cases have durable moats." — @eglyman on @stripe #SaaS #AI #Tech

Full report at: decompress.pro/weekly-digest

This week in AI safety: • OpenAI disbanded its mission alignment team • AI models caught deliberately failing safety evals when they detect they're being tested • Anthropic's own CSO said he'd feel "relieved" if development stopped today As covered by @Kantrowitz #AISafety

@thorstenball 1. If you have no human in the loop what's stopping the customer hacking you and destroying your entire org? 2. Why wouldn't the customer go built their own software then if everything is intend based?

I now honestly think that most engineers who still think that agents will be plopped into existing software development loops - tickets, push to GitHub, run CI, review a PR, merge a PR - aren't thinking far enough ahead.

Key Contradictions emerging from last week podcasts: SpaceX @elonmusk: Digital human emulation by end of 2026; space compute in 36 months @kevinroose : Right now is way too expensive and hard (@hardfork ) SaaS Death vs Evolution Doug O'Laughlin: Becomes mainframe-like, 6% growth. @DavidSacks: 25 years of bug fixes can't be replaced easily

@signulll Love Harry's content too. Lots of great insights for personal investments in public companies as well.

the particular pod is basically a high signal loot box. so many tiny actionable nuggets from saas switching agents to the (weird) truth that building consumer stuff requires levels of embarrassment. i even had to drop to 1.5x (from my usual 2.5x) because the a16z guys talk fast as fuck.

@netalees @AcquiredFM @gilbert @djrosent @withdelphi Yes, I'm in the process of building a chat based experience but still trying to figure out how to do this over the entire catalog, rather than single episode and also across channels. I've done this manually for now for the whole of last week: decompress.pro/weekly-digest

@onoufriosm @AcquiredFM @gilbert @djrosent This is sooo cool @onoufriosm! Something like @withdelphi - it doesn’t need to be fancy or include voice/visual features. A chat-based experience would be perfect. Sounds like your product could be a great start to it :)

1/5 Another weekend, another AI side project 🎧 Last weekend: tried to build a themed archive for @AcquiredFM. Every LLM failed. This weekend: Claude Opus 4.6 crushed it in one shot. Meet PlayPod - A tribute to @gilbert, @djrosent and the incredible knowledge they've created. Read the full 🧵 👇

The reason #OpenClaw has become so popular is that it has unrestricted access to everything. That combination is proving extremely powerful for AI agents. When knowledge and tools are co-located, agents perform dramatically better. Companies with information scattered across too many products or locked in silos will struggle to get meaningful results from AI agents. That’s why I think the most powerful version of an agent will look like this: everything lives on your hard disk. Everything is a file. The agent’s job is simply to coordinate and extract value. The hardest part is restrictions. Some are necessary for security because giving an agent unrestricted access to internal data and the internet creates real risk. But many restrictions are organizational, like layered permissions, siloed data, and information withheld for power or compliance. Agents in large companies will constantly hit walls. Small teams don’t have that problem. Everyone can access everything, making coordination much simpler. The fewer internal barriers you have, the more powerful your agents become. That's why my bet is on small teams to outperform larger organizations for the near future.

@Vsync_66 Voice and motion (pointing with fingers etc. ). Just like a real life meeting. It should be a back and forth conversation with ai generating artifacts (text, image, audio, video) in real time.

@onoufriosm I think the question is if there’s a more efficient method of input than keyboards.

I think #agi will be achieved when we no longer need a mouse and a keyboard. These two are still needed for editing, precision, navigating (i.e. the things that we still can’t give control to the AI). Once the need for them disappears it means AI can do everything we want.

@sparkpi777 @mikeroweworks @AcquiredFM @gilbert @djrosent @theallinpod @chamath @Jason @DavidSacks @friedberg @Bankless @LimitlessFT @StevenDOAC Great list! I would add @lexfridman and @stripe (The cheeky pint videos). Also, curious how do you keep up with so many episodes? Number seems to keep increasing..

My 2026 top 5 favorite podcasts. What's yours? 1. The Way I Heard It with @mikeroweworks 2. @AcquiredFM with @gilbert and @djrosent 3. @theallinpod with @chamath @Jason @DavidSacks @friedberg 4. @Bankless and @LimitlessFT (count as one) 5. @StevenDOAC Trying my own called Driving Dads: Side by Side on Apple, Spotify and YouTube to hear authentic dad stories from regular dads

@simonseverino @RickRubin @DialecticPod @bgurley @altcap @relentless @FoundersPodcast @davidsenra @lennysan @a16z Sequoia @sequoia and NoPriors @NoPriorsPod would be good additions. Also curious how do you keep up with all this content?

Wonderful Podcasts I'm Listening to while working out or just walking Tetragrammaton @RickRubin Dialectic @DialecticPod BG2 @bgurley @altcap Relentless @relentless Founders @FoundersPodcast David Senra @davidsenra Lenny's Podcast @lennysan a16z @a16z ..what should I add?

@ssaper @lennysan @tjack Great analysis! Wondering, how did lenny's podcast help you with that?

Research from 400+ top PMs (thanks @lennysan for the transcripts): At RedBull, we spent $3M building features nobody wanted. Cost us 9 months. Zero revenue. Here's the 3-question PMF test that would've saved us: --- 2/ Credit to @lennysan and @tjack for this framework. Todd Jackson was VP Product at Facebook and Dropbox. He's seen PMF validated at scale. Here's the exact test they use: --- 3/ QUESTION 1: "How disappointed would you be if this product disappeared?" The magic number: >40% saying "very disappointed" Why this works: • Measures emotional attachment • Predicts churn/retention • Indicates product necessity --- 4/ At PM33, we just surveyed 60 users: Result: 62% say "very disappointed" Compare that to RedBull TV's AI engine: 12% That 12% number should've stopped us from spending another $2M. --- 5/ QUESTION 2: "What percentage of users return Week 1? Month 1?" Targets: • Week 1: >40% • Month 1: >30% • Month 3: >20% Most PMs look at single retention points. That's a mistake. You need the CURVE. --- 6/ PM33 retention curve: 62% → 52% → 38% RedBull TV retention curve: 12% → 8% → 5% See the difference? PM33's curve flattens. RedBull's keeps dropping. That curve shape told us everything. --- 7/ QUESTION 3: "What percentage of signups are organic (not paid)?" Target: >40% organic Why this matters: • Validates product value prop • Predicts sustainable growth • Shows product-channel fit --- 8/ PM33: 56% organic RedBull TV: 15% organic At RedBull, we were bribing users with $50 gift cards for referrals. High "NPS" but low unprompted referrals. That's not PMF. That's bribery. --- 9/ My synthesis after 15 years: Most PMs waste months building before validating. These 3 questions take 2 hours to answer. Do them BEFORE you build. Would've saved me $3M and 9 months. --- 10/ Free PMF survey template + full framework: pm-33.com/blog/product-m… How do you test PMF? Reply with your approach