Friday, May 29, 2026
AI revenue run rates hit $200 billion as agents move from chat to decision traces
May 29 · 8 videos
OpenAI and Anthropic are scaling faster than any software companies in history.
Meta is spending $30 billion to pivot into an AI subscription powerhouse.
Hugging Face released a $300 hackable robot.
GigaML founders turned down $550k to build agents for DoorDash.
The token path is the new software budget.
“Anthropic and OpenAI are adding more revenue per month than Meta, Google or Microsoft.”
The forgotten developer who saved JavaScript...
Jeremy Ashkenas · Fireship · 6 min
Watch on YouTube →The modern JavaScript ecosystem was shaped by Jeremy Ashkenas through three pivotal tools: Underscore.js, CoffeeScript, and Backbone.js. These projects filled critical gaps in the language's original design and influenced the ECMAScript standards.
- JavaScript was designed in only 10 days, leading to a fragmented and broken language in its early years.
- Jeremy Ashkenas created Underscore.js with 60 helper functions to provide a standard library for developers.
- CoffeeScript introduced clean syntax like arrow functions and classes that were later absorbed into ES6.
- Backbone.js provided a 2,000 line MVC structure that proved client side heavy applications were possible.
- CoffeeScript's inclusion in Rails 3.1 led to immediate enterprise level adoption of these modern patterns.
- Successful tools often become obsolete because their best ideas are eventually integrated into the core platform.
Reachy Mini: the $300 open source robot you can actually hack — Andres Marafioti, Hugging Face
Andres Marafioti · AI Engineer · 21 min
Watch on YouTube →Andres Marafioti from Hugging Face introduces the Reachy Mini, a $300 hackable robot designed to democratize robotics research. The talk focuses on hardware repairability and optimizing voice AI for real time interaction.
- Hugging Face has shipped 7,500 Reachy Mini units to encourage creative interaction and user repairability.
- Humanoid forms are often less agile and stable than alternative biological inspirations like spiders.
- The TTS model was optimized from 0.8x real time to 5.8x real time with under 200ms latency.
- Switching from dynamic to static KV caches enabled CUDA graph captures for significant acceleration.
- The architecture uses Parakeet for speech to text and Qwen 3.5 27B for the LLM logic.
- Shipping products unassembled creates a community of users capable of modifying their own hardware.
Biggest Mysteries in Physics: Antimatter, Dark Energy & ToE - Don Lincoln | Lex Fridman Podcast #497
Don Lincoln · Lex Fridman · 173 min
Watch on YouTube →Physicist Don Lincoln discusses the history of unification in physics and the massive gaps remaining in the Standard Model. He explores why the next revolution in understanding reality may take centuries to validate.
- There is a 10^120 discrepancy between the theoretical and observed vacuum energy of empty space.
- Gravity likely unifies with other forces at an energy scale a quadrillion times higher than the LHC can reach.
- The world production rate of antimatter is only about 1 nanogram per year.
- Scientific progress is driven by empirical anomalies like the rotation of galaxies rather than pure theoretical elegance.
- The discovery of the Higgs Boson was a major milestone, but the Standard Model remains fundamentally incomplete.
- Fundamental research into subatomic forces typically takes over 100 years to yield technological spin offs.
Why your agents need decision traces, not just documents — Zach Blumenfeld, Neo4j
Zach Blumenfeld · AI Engineer · 20 min
Watch on YouTube →Zach Blumenfeld argues that autonomous agents require Context Graphs to record the reasoning and causal chains behind past actions. This shift moves AI from simple question answering to complex organizational decision making.
- Context Graphs incorporate decision traces which include the reasoning and precedents behind past actions.
- Agents can surface past decision paths using a combination of vector search and graph embeddings.
- Neo4j released a uvx command to instantly scaffold full stack context graph applications.
- The neo4j-agent-memory package handles entity extraction, deduplication, and multi layered memory.
- Explainability is improved by pointing to specific past precedents that justify a current recommendation.
- The tool includes 22 pre-built domains to lower the barrier to entry for graph based AI.
It's Time to Stop Ignoring Meta
Josh Kale · Limitless Podcast · 23 min
Watch on YouTube →Meta is pivoting to become an AI subscription powerhouse with a $30 billion CapEx commitment. The discussion covers Meta's ecosystem strategy and the rise of orbital data centers.
- Meta is launching tiered subscriptions for WhatsApp and Instagram starting as low as $2.99.
- SpaceX and StarCloud are collaborating on orbital data centers to target 5 cents per kilowatt power costs.
- Snowflake's stock jumped 30% as it positioned itself as an orchestration layer for multiple frontier models.
- Meta's deep integration into social attention cycles provides a data advantage competitors cannot replicate.
- The NeoCloud model involves renting out massive compute clusters if proprietary models fail to dominate.
- Vibes based usage is often a more accurate indicator of model strength than formal benchmarks.
The New Rule for Picking AI Winners | The a16z Show
David George · a16z · 33 min
Watch on YouTube →David George and David Clark discuss the tectonic shift in venture capital as AI startups scale revenue at unprecedented speeds. They define the new strategic imperatives for capturing value in the AI cycle.
- OpenAI and Anthropic are projected to reach a combined $200 billion run rate by the end of 2024.
- The threshold for a top 1% venture exit has risen from $10 billion to $32 billion in 24 months.
- Approximately 40% of top AI startups drop off the leaderboard annually due to high market velocity.
- Companies must be in the token path to capture spend being diverted from legacy software budgets.
- Physical infrastructure like power and data centers is the primary governor of growth in this cycle.
- AI technology has currently achieved less than 5% diffusion into the broader real economy.
You Aren’t Crazy… Something Is Wrong With People Nowadays
Rob Dial · The Mindset Mentor Podcast · 16 min
Watch on YouTube →Rob Dial explores the mismatch between ancient human biology and modern technological growth. He argues that deep focus and emotional regulation are becoming the most valuable competitive assets.
- The human brain structure is 200,000 years old and cannot distinguish digital notifications from physical threats.
- Modern humans consume more data by 9:00 AM than their ancestors did in several weeks.
- Chronic anxiety is often a manifestation of the nervous system rejecting modern living conditions.
- Silence is necessary to reveal unresolved emotions and fears that are typically avoided through distraction.
- Rare traits like independent thinking and deep focus will be high value advantages in the next decade.
- The brain is 95% to 99% identical to that of early Homo Sapiens.
Why Two IIT Engineers Turned Down $550K Jobs To Build A Startup
Varun Vummadi · Y Combinator · 24 min
Watch on YouTube →Varun Vummadi of GigaML shares how his team pivoted from a rejected YC idea to building high performance AI agents for enterprise clients. The company focuses on technical excellence over large sales teams.
- Varun Vummadi turned down a $550,000 quant job offer to pursue his startup GigaML.
- GigaML achieved support deflection rates of 60-70% for major clients like DoorDash.
- The team uses a Markdown to KPI framework to treat AI agent behavior as an iterative configuration problem.
- A small team of 8 people uses coding agents to maintain a 7x productivity multiplier.
- Charging customers early is the best way to validate if a startup is solving a real problem.
- The company landed Fortune 500 clients by leveraging the YC network and technical merit.
References
PeopleJeremy Ashkenas · Andres Marafioti · Don Lincoln · Zach Blumenfeld · David George (x.com/DavidGeorge83) · David Clark (x.com/daveclark85) · Varun Vummadi · Rob Dial · Mark Zuckerberg · Elon Musk · Brendan Eich · David Heinemeier Hansson
ToolsReachy Mini · Neo4j · GigaML · Underscore.js · CoffeeScript · Backbone.js · Parakeet · Qwen 3.5 · Coqui TTS · Snowflake · Cursor