Alex Lieberman

I'm trying something crazy. I'm going to pay a writer like a salesperson. You see, part of my plan to make @tenex_labs the McKinsey of AI is to build a worldclass media company on top of it. A media company that helps knowledge workers stay on the right side of a post-AI economy. One key piece of the plan is hiring an AI-obsessed, informed yet punchy writer to OWN our newsletter, Ultrathink, and our long-form website content (think AI playbooks, reports, research). But here's the kicker... I don't just want you to make words dance off the page & smack the reader across the face with value. I want you to drive customers for the mothership, Tenex engineering & AI transformation. Which is why, I'm going to incentivize you to do so. To the person selected as our writer, not only will you be the keeper of our written content, but you'll also participate in the $$$ driven by your content. Do I have your attention? Apply below.

We founded @MorningBrew on a set of principles. Those principles have evolved, but they've never been more important. In a post-AI world, trusted distribution is one of the final moats. Here are the 12 rules of the media game as I see them today: Rule #1: Personality-driven content is becoming a bigger and bigger part of consumers’ media diets. - The democratization of creation & distribution (read: internet & social platforms) means that people can attract as much distribution as institutions. Consumers have always longed for content that feels personal & relatable. Rule #2: Brand-driven content isn’t dead. The expectations of it have just changed & it serves a more important role in media than ever before. - In an era of personality-driven media, consumers expect relatability & resonance with brands, even if they’re faceless. This just means that even if a person isn’t attached to the brand, consumers want to feel an individual person behind it. - Personality-driven content is a necessity for most media brands today, but with it comes a lot of risk. One of the ways to derisk personality risk is by having faceless franchises that give you ownership over a healthy % of your audience. Rule #3: Riches are in the niches. - Democratization of creation & distribution not only increased consumer appetite for personality-driven content. It also drove a content supply glut, allowing consumers to be pickier than ever before. - We went from 12 channels on the original TVs to near unlimited choice. And in a world of unlimited choice, consumers can find great content related to any/all interests from the most mainstream to the most obscure. Rule #4: Diversifying risk across the audience funnel is key. - There are three types of audiences: rented audience, owned audience, and monetized audience. Each type of audience offers benefits & trade-offs to a company, so it’s less about targeting a single type and more about maximizing the benefits of each, while mitigating the risks. - Rented audiences offer rapid growth, but platform risk. - Owned audiences put you in control of your audience, but growth is slower. - And monetized audience is how you pay the bills, but you can’t monetize your audience without building trust via rented & owned. Rule #5: Content channels & trends change but human psychology does not. - Human beings are selfish. We selfishly do things that satisfy our most basic desire: survival. Survival in modern day society boils down to physical safety & psychological safety. - Physical safety: For the modern day media consumer, that looks like acquiring knowledge that helps you: make money, improve professionally (so you can make more money), or save money. - Psychological safety: For the modern day media consumer, that means consuming content that makes you feel belonging (i.e. entertainment, hobbies, niches, passion areas) and helps you with courtship (i.e. attracting a partner/procreating). Rule #6: We are late stage on most content platforms. - In the early days of Instagram, Tiktok, and Twitter, attention was cheap, and you could amass a large audience simply by being an early adopter and actively creating on-platform. Those days are (mostly) gone. Aside from LLMEO/GEO, you can no longer build distribution simply by taking at-bats. - As platforms have entered the later innings of their life cycles, content has gone from scarce to abundant, while attention has remained finite. This means that the quality bar for what is “worthy” of a consumer’s time continues to increase. Rule #7: The tiktok-ification of media is almost complete. - Social platforms were built on the social graph. You see the content of the people you follow and your content is seen by people that follow you. - Practically speaking, that meant those with big audiences would always get more views than those with small audiences. - Interest graphs (made popular by TikTok) flipped internet media on its head. With the advent of For You, you now see content the platform believes you’d be interested in, and your content is seen by people the platform thinks would be interested in your content. - In practice, this means that every creator/brand is only as successful as its last post, the variance of performance has never been higher, and it’s never been easier for someone with a small audience to get outsized results on a single piece of content. Rule #8: Software is becoming content. - As the barriers to build software approach 0, software is a new content medium which can be used to build and own audience. Rule #9: AI is a supporting actor in media. - The question to ask is not how much of this content is AI or human-generated. - It’s how well does this content resonate with its audience and accomplish one of the base human needs. - Human judgement and taste is still essential in accomplishing that goal no matter how much or little AI is used. Rule #10: No content is original. - It’s all a remix of prior ideas that came before. Great artists steal, bad artists copy. Media companies should behave in the same manner. Rule #11: A modern media company serves 3 customers - Its audience, its creators, and those who pay the bills. - The product a media company offers to serve each of these customers can be/is likely different. Rule #12: A media company is like an investor. - Deploying resources based on the risk/return profile of an investment and the amount of resource required. - Example: a book should start as a blog. A blog should start as an article. An article should start as a tweet.

I want to start a community dedicated to Claude Code. It’s become the gateway drug to coding and experiencing the power of AI for tons of people. This will be a space for people to share killer use cases, agentic workflows, proven prompts, and connect with other CC obsessives. Comment “Claude” if you want to join.

If you’re looking for a “job” don’t join an early-stage startup. If you’re looking for a smorgasbord of growth, heartbreak, meaning, best friends, and some money, definitely join an early-stage startup.

We can't hire fast enough at @tenex_labs. We're building McKinsey for AI, growing from 15 to 150 this year, and working with some of the biggest companies in the world. 1) ai engineers: fullstack, cracked, ai-native, nyc-based (or open to relocation), paid like a salesperson (base + uncapped variable upside), startup experience preferred. 2) newsletter writer: own tenex's flagship AI newsletter, lead creation of long-form content (AI playbooks + reports); all-in on using AI to increase leverage; deeply interested about applied AI. 3) technical acquisition lead: we must hire 135 people this year. that is absurd. we must trust you to own that end-to-end. JDs & applications below

I love @elonmusk & think @nikitabier is sharp af, but I don't understand why X algo was open-sourced. 1) Visibility makes clear rules of the game 2) Bad actors will pounce on gaming the game 3) Bad actors don't care about creating great content 4) Content gets worse 5) Algo needs to change to account for the game being gamed 6) Spiral continues I must be thinking about something incorrectly.

Software and content have never been more similar. 1) Cost of creation is approaching zero. 2) Attention is finite, but supply has skyrocketed. 3) Trusted distribution has never been more important. 4) Messy middle of mediocrity has 10x'd in size. 5) Curators & tastemakers are high status. 6) Expansion of choice means customers have higher standards & expect customization. 7) Toy software becomes marketing vs. utility for businesses. 8) Explosion of indie software creators like indie content creators.

The average CEO cannot tell you the difference between an automation and an AI agent. This breakdown (h/t @wadefoster ) makes it glaringly obvious. An automation is anything that requires no-trust decision making (if this, then that). An AI agent is anything that requires some-trust decision making. The most powerful workflows that I’m seeing businesses use today are neither pure automations nor pure agents. They’re agentic workflows. Agentic workflows get the leverage of AI (read: intelligent decision making) that deterministic software never offered. But they also have the predictability of automations, where the cost of error is too high. Knowledge work will continue to get pushed further right on this spectrum as the technology improves, but today living in the middle is often the sweet spot.

I'm having the most fun i've ever had in my career right now. The AI overlords might take over 5 years from now, but for right now, I'm happy as hell.

I need to hire 60 engineers this year. I want to build an AI agent/orchestration of agents that acts as a world-class sourcing recruiter to make this possible. High-level design: 1) Agent scours a recruiting pool (think producthunt, glassdoor, indiehackers, X, linkedin, github) and identifies prospects of interest based on qualitative/quantitative criteria 2) Agent adds prospects of interest to a database, which then gets enriched with Linkedin URL, email, and anything else that is necessary to send a customized message 3) Agent runs an outreach campaign to candidate (with combination of messaging & outreach channels) from my accounts (my email and linkedin). If candidate expresses interest in learning more, we send them our tik-tok style hiring site with more info and if they're still interested they formally apply. 4) Agent tracks their progress in Ashby (our ATS) and then updates its list-building strategies/focus based on which sources are responsible for the most late-stage/hired candidates. How should I approach building this? P.S. if you are a cracked engineer who wants to solve hard problems with brilliant people in NYC, apply for a job at tenex[.]co/about-us P.P.S. if you're a recruiter trying to win my business, please don't waste your time. will reach out to external recruiters on my own time.

I’m non-technical but want to deeply understand AI. @karpathy's “Intro to LLMs” is the best resource I’ve found so far. Here are my biggest takeaways and questions from his 60-minute talk: 1. A large language model is “just two files.” Under the hood, an LLM like LLaMA‑2‑70B is literally (1) a giant parameters file (the learned weights) and (2) a small run file (code that implements the neural net and feeds data through it). Question: If the architecture code is tiny and public, what actual moat is left besides the weights? 2. Open‑weights vs closed models. LLaMA‑2 is open‑weights: architecture + weights + paper are public. GPT‑4, Claude, etc. are closed: you get an API/web UI but not the actual model. Question: For a company, when is “renting” a closed model strategically worse than owning an open‑weights model? 3. Training vs inference: training is the hard, expensive part. Running the model (inference) is cheap; getting the weights (training) is a major industrial process. Question: Where is the greatest axis of innovation in front of us to lower the cost of training significantly? 4. Pre‑training compresses ~10 TB of internet text. LLaMA‑2‑70B is trained on roughly 10 TB of scraped internet text, compressed into 140 GB of parameters—a ~100× lossy compression of “internet knowledge.” Question: Given that we’ve run out of knowledge on the internet to pre-train models on, is new data going to be the limiting factor on model improvement moving forward? 5. Training scale: ~6,000 GPUs × 12 days ≈ ~$2M for LLaMA‑2‑70B. That’s already described as “rookie numbers” compared to modern frontier models, which are ~10× bigger in data/compute and cost tens to hundreds of millions. Question: How far are we from “more compute” no longer being a competitive advantage? 6. Frontier models just scale this up by another ~10×. State‑of‑the‑art models (i.e. GPT‑5) simply dial up parameters, data, and compute by large factors relative to LLaMA‑2‑70B. Question: How much of GPT‑5‑style capability is just more scale vs genuinely new algorithms? 7. Core objective of an LLM predict the next word in a sequence. LLMs are trained to take a sequence like “the cat sat on the” and predict the probability distribution over the next word (“mat” with ~97%, etc.). Question: The beauty and the curse of LLMs is them being probabilistic. How can we create the right constraints such that people trust LLMs in enterprise settings? 8. Architecture is known: the Transformer. We know all the math and wiring (layers, attention, etc.); that part is transparent and simple relative to behavior. Question: If the architecture is commoditized, where exactly do you build sustainable differentiation? And how much more shelf life is there on the Transformer before a new architecture takes over? 9. Parameters are a black box. Billions of weights cooperate to solve next‑word prediction, but we don’t really know “what each one does”—only how to adjust them to lower loss. Rabbit hole: Read about mechanistic interpretability work. 10. Treat LLMs as empirical artifacts, not engineered machines. They’re less like cars (fully understood mechanisms) and more like organisms we poke, test, benchmark, and characterize behaviorally. Rabbit hole: Understand the current process for evals & if/what limitations exist in today’s eval tools. 11. Pre‑training vs. fine-tuning. Pre-training favors quantity over quality; Fine-tuning flips that: maybe ~100k really good dialogs matter more than another terabyte of web junk. Question: How much incremental performance can fine tuning and RHLF drive for models? Is it a fraction of what pre training does for performance or is it more meaningful than that? 12. Knowledge vs behavior. Pre-training loads the model with world knowledge; Fine-tuning teaches it to be helpful, harmless, and to respond in Q&A format. Rabbit hole: I’d love to deeply understand how exactly a model is fine tuned from beginning to end. 13. Reinforcement learning from human feedback (RLHF) via comparisons. It’s often easier for labelers to rank several options vs. write the best one from scratch; RLHF uses these rankings to further improve the model. Question: When exactly does it make sense to fine tune a model vs. use RHLF & does the answer depend on the domain of knowledge the model will be used for? 14. Closed vs open models. Closed models are stronger but opaque; open‑weights models are weaker but hackable, fine‑tunable, and deployable on your own infra. Question: As companies deploy agents, what is the most important consideration to make as they think about their AI tech stack? 15. Scaling laws: performance is a smooth, predictable function of model size and data. Given parameters (N) and data (D), you can predict next‑token accuracy with surprising reliability, and the curve hasn’t obviously saturated yet. Question: If capabilities keep scaling smoothly, what non‑technical bottlenecks (data rights, energy, chips, regulation) become the real limiters? 16. GPU and data “gold rush” is driven by scaling law confidence. Since everyone believes “more compute → better model,” there’s a race to grab GPUs, data, and money. Question: Let’s assume scaling laws no longer scale. Who is most screwed when the music stops? 17. LLMs as tool-using agents, not just text predictors. Modern LLMs don’t just “think in text”; they orchestrate tools. Given a natural-language task, the model decides to (1) browse the web, (2) call a calculator or write Python to compute ratios and extrapolations, (3) generate plots with matplotlib, and (4) even hand off to an image model (like DALL·E) to create visuals. The intelligence is increasingly in the coordination layer: the LLM becomes a kind of “foreman” that plans, calls tools, checks outputs, and weaves everything back into a coherent answer. 18. How do LLMs know when to make a tool call? “It emits special words, e.g. |BROWSER|. It captures the output that follows, sends it off to a tool, comes back with the result and continues the generation. How does the LLM know to emit these special words? Finetuning datasets teach it how and when to browse, by example.” 19. System 1 vs System 2 thinking applied to LLMs. Concept popularized in Thinking Fast and Slow. System 1 = fast, instinctive; System 2 = slower, deliberate, tree‑searchy reasoning. Right now LLMs mostly operate in System 1 mode: same “chunk time” per token. Rabbit hole: Explore how “chain‑of‑thought” method works & what limitations still exist in System 2 thinking for LLMs. 20. Desired future: trade time for accuracy. This was before the first reasoning model (GPT O1) came out. At the time, Karpathy talked about this idea of wanting to be able to say: “Here’s a hard problem, take 30 minutes,” and get a more accurate answer than a quick reply; currently, the models can’t do that in a principled way. 21. Model self‑improvement example: AlphaGo’s two stages. AlphaGo first imitates human Go games, then surpasses humans via self‑play and a simple, cheap reward signal (did you win?). Question: What’s the best way to improve models in domains where there isn’t a simple reward function, like creative writing or design? 22. Retrieval‑augmented generation (RAG) as “local browsing.” Instead of searching the internet, the model searches your uploaded files and pulls snippets into its context before answering. Question: Where does RAG break down in production? 23. Think of LLMs as the kernel process of a new operating system. This process is coordinating resources including tools, memory, and I/O for problem-solving. Future LLM will: - read/generate text - have more knowledge than any single human about all subjects - browse the internet - use existing software infrastructure - see and generate images and video - hear and speak and generate music - think for a long time using system 2 - “self-improve” in domains with a reward function - customized and fine-tuned - communicate with other LLMs Rabbit hole: Draw out the LLM OS and explain it to someone. This will show how well you understand the technology. 24. The LLM OS is reminiscent of today’s operating systems. The finite context window is like working memory; browsing/RAG are like paging data in from disk or the internet; rapidly growing closed vs. open ecosystem; Managing what’s in context is a core challenge. Rabbit hole: Explore techniques for working across many context windows & longer-running tasks. 25. New computing stack → new security problems. Just as OS’ created new attack surfaces (malware, exploits), LLM‑centric stacks create their own families of attacks. Jailbreaks, adversarial prompting, adversarial suffixes, and prompt injection. Question: security for AI systems seems orders of magnitude harder than traditional software because the # of edge cases feels infinite. Is this assumption right or wrong? 26: LLMs are a new computing paradigm with huge promise and serious challenges. They compress internet‑scale knowledge, act as operating‑system‑like kernels, orchestrate tools and modalities, and open up both transformative products and novel security risks. Question: what is the most nascent part of the LLM OS that needs to be built up in order to accelerate diffusion of the technology? Link to the full “Intro to LLMs” video below

I've always wondered why more podcasts don't record live to build an obsessive audience. So I decided for my (upcoming) AI podcast, Human In The Loop, every episode will start with a live recording followed by posting the edited version. Example: Today, I'm having @wadefoster teach me how to build AI agents in @zapier . (RSVP below) 600+ people are already registered for the live event. This does two things: 1) Grows our email list wayyy faster 2) Allows people to engage with badass guests Post-event, we'll repurpose this anchor content and turn it into the following: - Long-form YT video - Long-form podcast - On-demand version of the video on website - Tactical playbook on website - Hero story of our (upcoming) newsletter - Short-form video cuts (for IG, TT, LI, YT) - Text-based social posts with highlights This is how you build community & stretch one piece of great content as much as possible.