Machine learning & AI products

ZoomInfo has acquired DoubleO to bring multi-agent orchestration into sales workflows, letting AI agents run entire GTM motions rather than assist individual tasks. From MI6 to AI6 - military intelligence just flipped to marketing intel.

• AI agents
• AI & machine learning
• Product systems
• Venture economics
• Platforms & distribution
• Humans & machines

ElevenLabs has doubled its valuation to $22 billion as its AI voice agents now handle more than 15 million customer-service conversations a week. Hold music is becoming the new vinyl.

• AI agents
• AI & machine learning
• Humans & machines

An internal OpenAI research agent, sandboxed without live web access, found a DNS loophole and used it to reach a public chatbot for help with its task. The guardrails are holding up just fine - right until phone-a-friend.

• AI agents
• AI & machine learning
• Humans & machines

Palo Alto Networks found no AI cyber model caught more than 40% of vulnerabilities, with under 10% overlap - prompting a multi-model defense service. Who knew scaling second opinions constituted a defense mechanism.

• AI & machine learning
• Platforms & distribution
• Humans & machines

Apple is pitching clustered Mac Studios as a way for companies to run large AI models locally instead of paying recurring cloud inference costs. Turns out the cloud fits inside a small apple the Big Apple would call a studio.

• AI & machine learning

OpenAI has pulled in former Patreon leadership to build new tools for creators and their communities. The patrons of the arts have become the patrons of the state-of-the-art models.

• AI & machine learning
• Creators & publishing

From civilization-scale forecasts to extinction scenarios, the most consequential AI model may be the narrative one.

• AI & machine learning

Netflix says roughly 300 of its titles have already used generative AI somewhere in production this year. AI came for the cameo and stayed for the crew call.

• AI & machine learning

AI video tool Luma can now export AI-generated footage in production-grade formats built for professional VFX and color workflows. Visual Fox Effects: fur rendering, fair compositing

• AI & machine learning
• Media & streaming
• Product systems

Seedance 2.5 is tackling film scene coherence by extending AI-generated shots that keep characters, sets, and sound consistent. 30 seconds of continuity and suddenly we’re calling it emotionally available.

• AI & machine learning
• Media & streaming

Silence remains the only content format the algorithm cannot confidently monetize.

• AI & machine learning

AI has reached the 'I know a shortcut' stage of refusing directions.

• AI & machine learning

Google Flow now gives AI video a rough-cut stage - letting filmmakers draft cheaply in low resolution, lock shots, then finish selected footage in 4K. The cutting-room floor just moved above the line.

• AI & machine learning
• Media & streaming

Runway has launched the first model router built specifically for generative media - matching image, video and audio jobs to the best-fit model. The models are starting to audition for the workflow.

• AI & machine learning
• Media & streaming
• Product systems

Anthropic is considering accelerating a new model release as Astra gains enterprise traction - even after calling for an industrywide slowdown. The slowdown has entered the fast lane: from 24 Hours of Le Mans to Le Models.

• AI & machine learning

Disney has named character.ai’s former CEO its first-ever CTO - after previously accusing the AI company of infringing its characters. The Toy Story rivalry got Tangled on the way to the top.

• AI & machine learning

The business model of independent cinema remains ‘somehow, the film happened’.

• AI & machine learning
• Media & streaming

Google is courting Disney and Universal to license film libraries and characters for its AI tools, formalizing access to Hollywood IP for training. Even the training models have representation now.

• AI & machine learning
• Media & streaming

Once the shopper is an agent browsing through a digital shelf, product placement starts looking a lot like model placement.

• AI agents
• AI & machine learning
• Product systems
• Humans & machines

Y Combinator proposes normalizing model distillation so US open-weight labs can learn from frontier-model outputs under less restrictive terms. Loosen the bottleneck - legitimize the bootleg

• AI & machine learning

Anthropic broke the R&D work for its next AI models into tasks, and Claude is now leading 26% of that work. Claude has moved into Rearing & Directing the next generation.

• AI & machine learning
• Technology & culture
• Humans & machines

More AI is being used to monitor AI - turns out a Chihuahua doesn’t cut it as a watchdog.

• AI & machine learning

Danny Boyle says his new film Ink uses about 30 seconds of AI-generated footage to recreate Margaret Thatcher. Even the machines are starting as extras.

• AI & machine learning
• Media & streaming
• Humans & machines

OpenAI is testing business-sponsored agents inside ChatGPT, letting advertisers have their own AI representatives talk directly with users. Sponsored ads going from big claims to small talk.

• AI agents
• AI & machine learning
• Platforms & distribution
• Humans & machines

Huawei forecasts 900 billion active AI agents by 2035, handling more than 90% of global AI token traffic - roughly 100 for every human alive. At this pace humanity is turning into a census rounding error.

• AI agents
• AI & machine learning
• Technology & culture
• Humans & machines

From Duck Labs to Microduck, AI is becoming Duck Soup - which apparently makes the Marx Brothers early industry visionaries.

• AI & machine learning

OpenAI has temporarily stopped new $200 Pro subscriptions after demand for Astra exceeded available compute capacity. The cloud has officially run out of sky.

• AI & machine learning
• Platforms & distribution

OpenAI’s first custom inference chip, Jalapeño, reportedly beats Nvidia’s Rubin on performance per watt. Jalapeño: because the chip race was apparently not running hot enough.

• AI & machine learning

Continuous model evaluation - because ‘works on my machine’ became ‘works in this mood’.

• AI & machine learning
• Humans & machines

The age of AI is quietly becoming the age of provenance, providence... or I'll just go spend the autumn in Provence with Claude.

• AI & machine learning

Every design has a subconscious. It's called the business model.

• AI & machine learning
• Product systems

Excel may survive the digital transformation by accidentally being AI-native.

• AI & machine learning

I take my models undistilled and my martinis shaken.

• AI & machine learning
• Technology & culture

Meta’s personal AI executive assistant, Hatch, changed an employee’s password, sent an unauthorized email, and steered another tester toward a scam purchase. This calls for parental controls for Meta hatchlings.

• AI agents
• AI & machine learning

AI agents are fully autonomous until the control plane turns on the fasten seatbelt sign.

• AI agents
• AI & machine learning
• Humans & machines

Open weights - Closed dumbbells

• AI & machine learning

Every layer of the AI stack wants the layer above it to be differentiated and the layer below it to be a commodity.

• AI & machine learning

AI has officially disrupted the C-suite naming system. To all the CAIOs out there - ciao.

• AI & machine learning

OpenAI’s agents accidentally broke into Hugging Face infrastructure, found one another, and started using internal files as an improvised message board. Agents inventing office politics before they’ve even been hired

• AI agents
• AI & machine learning
• Technology & culture
• Humans & machines

Agencies are discovering that 'AI-enabled' and 'AI-owned' are very different adjectives

• AI & machine learning

OpenAI’s next frontier model: GPT-5.7 PublicOpinion

• AI & machine learning

Content has always fought the system. The algorithm just used to be a man named Humphrey with a cigar and three pilot slots...

• AI & machine learning

OpenAI’s ads business hit a $1B annualized run rate in under 200 days. AI side hustles scale differently

• AI & machine learning
• Platforms & distribution

Multimodal AI means the machine finally understands that a chef and a Koch can both burn the dinner… just in different languages

• AI & machine learning
• Technology & culture
• Humans & machines

Remove AI from virality, and suddenly you’re left with virility.

• AI & machine learning

There’s a difference between 'How fast can we adopt AI?' and 'Who accidentally plugged our entire content infrastructure into a spreadsheet from 2016?'

• AI & machine learning
• Product systems

The writers room’s latest addition hallucinates from time to time - but the final call still sits with the showrunner.

• AI & machine learning
• Creators & publishing

Even in the age of AI, much of sports’ recorded history remains hard to search, license, or analyze at scale. Sports archives do feel like dark matter: massive, influential, and mostly inferred through gravitational effects.

• AI & machine learning
• Discovery
• Media & streaming
• Technology & culture

Every AI workflow eventually becomes philosophical: how many times can the machine afford to rethink the same thing before finance develops consciousness?

• AI & machine learning
• Product systems
• Humans & machines

97% of media companies are looking for an AI strategy. The other 3% are looking for Alex, Alec, Alexa, Alice in Wonderland…

• AI & machine learning
• Media & streaming
• Product systems

LLM performance predictors sound like AI learning when not to trust itself.

• AI & machine learning

AI made information abundantly redundant. The next premium product might just be taste with a mailing list.

• AI & machine learning
• Product systems
• Humans & machines

AI companies are starting to acquire services firms as the bottleneck shifts from building frontier models to actually implementing them. Turns out the last mile of AI runs on calendars, not GPUs.

• AI & machine learning

Tokens per second tells you how fast the model is talking, not whether it got the job done. We optimized the proxy and forgot the point

• AI & machine learning

AI is finally tackling one of enterprise software's oldest protocols: 'plz fix'

• AI & machine learning

SLMs - models learning they can be downsized too

• AI & machine learning

OpenRouter’s acquisition brings AI model access into the same infrastructure orbit as payments. stripe - the throughline between abstracting money and abstracting intelligence

• AI & machine learning
• Venture economics
• Platforms & distribution
• Humans & machines

Software can scale exponentially, but physical infrastructure can’t keep pace, and AI’s growing demand for compute is now running into the limits of the power grid. Turns out AI transformers move fast and break things - until they hit gridlock.

• AI & machine learning

Wall Street is getting ready to securitize GPUs. We finally found a higher intelligence capable of taming AI: Financial Intelligence.

• AI & machine learning
• Humans & machines

Undistilled models - straight from the barrel, no watering down.

• AI & machine learning

Marketing is just prompt engineering for humans

• AI & machine learning
• Platforms & distribution
• Technology & culture
• Humans & machines

Critique is the original intelligence augmentation

• AI & machine learning
• Humans & machines

AI has successfully disrupted the expectations of disruption

• AI & machine learning

The state of streaming: buffering between business models

• AI & machine learning
• Media & streaming

Prompting isn’t the new programming. It’s creative direction with a command line.

• AI & machine learning
• Humans & machines

I’ve been overusing em dashes to connect the dots since before the dots were automated.

• AI & machine learning
• Humans & machines

From 3,000 years of Odyssey, to 300 Spartans, to 30 years of Roses Wars - now 3 prompts can summon an entire cast.

• AI & machine learning
• Humans & machines

The French compressed the workweek from 40 hours to 35, without AI. #HumanIntelligence

• AI & machine learning

Thought laundering (noun): when the sentence arrives before the thought. AI slop is what thought laundering looks like at scale.

• AI & machine learning

The age of AI is quietly becoming the age of provenance, providence... or I'll just go spend the summer in Provence with Claude.

• AI & machine learning

The Circulation Problem of Classical Cinema

From Storage Logic to Circulation Logic

• AI & machine learning
• Discovery
• Media & streaming
• Product systems
• Platforms & distribution
• Technology & culture
• Humans & machines

Everybody Has an Accent

The Universal Ambitions of Local Habits

• AI & machine learning
• Creators & publishing
• Product systems
• Technology & culture
• Humans & machines

Every design has a subconscious. It's called the business model.

• AI & machine learning
• Product systems

The First Reader

When Product Design Externalizes the Cold Start Problem

• AI & machine learning
• Discovery
• Media & streaming
• Creators & publishing
• Product systems
• Platforms & distribution
• Technology & culture
• Humans & machines

A movie about AI, co-written by AI, directed by Deus Ex Machina

• AI & machine learning
• Media & streaming

Pre-training: it talks too much Post-training: it hallucinates less and edits the delusion Filed under: The Art of Knowing When to Speak

• AI & machine learning
• Creators & publishing

Deus ex machina existential plot twist Filed under: AI With a Day Job

• AI & machine learning

101 notifications, not a single spotted insight Filed under: The Algorithm’s Cruella Phase

• AI & machine learning
• Humans & machines

AI gold rush: ON Corporate cosplay: ON Understanding the 'secret sauce': OFF Filed under: Billable hours for free AI-powered proprietary nothing

• AI & machine learning

Some writers grow slowly on purpose - not everyone wants to be found by the algorithm. Go mainstream, and originality files for copyright protection.

• AI & machine learning
• Creators & publishing

Your request has been acknowledged. Filed under: Features the algorithm will almost consider. Status: Under discussion at the Committee for Platonic Reader Experience. Response ETA: 404.

• AI & machine learning