Pentagon pressures Anthropic on AI & Open source in age of cloning - Tech News (Feb 27, 2026)
February 27, 202613m 17s
Audio is streamed directly from the publisher (mcdn.podbean.com) as published in their RSS feed. Play Podcasts does not host this file. Rights-holders can request removal through the copyright & takedown page.
Show Notes
Please support this podcast by checking out our sponsors:
- Effortless AI design for presentations, websites, and more with Gamma - https://try.gamma.app/tad
- KrispCall: Agentic Cloud Telephony - https://try.krispcall.com/tad
- Invest Like the Pros with StockMVP - https://www.stock-mvp.com/?via=ron
Support The Automated Daily directly:
Buy me a coffee: https://buymeacoffee.com/theautomateddaily
Today's topics: Pentagon pressures Anthropic on AI - The U.S. Defense Department is pressuring Anthropic to allow Claude for “all lawful” military uses, with threats including contract termination and “supply chain risk” labeling—raising major AI governance and safety-limit questions. Open source in age of cloning - Ghost founder John O’Nolan argues AI is undermining open source’s assumptions by making rapid re-implementation and reverse engineering easier, blurring whether licensing still protects maintainers when models can rebuild systems “from scratch.” AI science, math, and labs - OpenAI’s Kevin Weil described a near-term shift toward closed-loop discovery—models designing experiments, robotic labs running them 24/7, and results feeding back via reinforcement-like cycles—while new AI-assisted math proofs keep piling up under scrutiny from experts like Terence Tao. AI competition, adoption, jobs - Jason Furman says frontier AI is unusually competitive with easy model switching and rapid price deflation, while Citadel Securities’ Frank Flight argues adoption is still S-curve-shaped—so mass white-collar displacement isn’t showing up in labor data yet. Chip wars: TPUs, High-NA - Meta is reportedly renting Google TPUs in a multi-year deal as alternatives to Nvidia expand, DeepSeek is said to be optimizing first for Chinese chip suppliers, and ASML says High-NA EUV tools are ready for high-volume manufacturing—each reshaping compute supply. What AI coding tools recommend - A large benchmark of Claude Code’s tool choices suggests it often “builds, not buys,” frequently selecting Custom/DIY solutions while still strongly defaulting to certain third-party picks like GitHub Actions and Stripe—quietly standardizing stacks. Agents: copilot versus autopilot - Kenneth Auchenberg frames AI agents as either “Leverage” tools that multiply a person’s output or “Function” tools that replace a job end-to-end, a distinction that affects pricing, go-to-market, and the path from copilot to autopilot. Tesla Cybercab leadership turnover - Tesla’s Cybercab program manager Victor Nechita is departing during the ramp toward volume production, highlighting continued leadership churn as Tesla pivots harder toward autonomy, robotaxis, and AI-driven strategy. Social media addiction lawsuit trial - A bellwether trial in Los Angeles features testimony from a young woman who says social media addiction began in childhood, as plaintiffs argue features like infinite scroll and autoplay harmed youth mental health—claims Meta and YouTube deny. Geothermal power and lithium extraction - The UK switched on its first deep geothermal electricity plant at United Downs in Cornwall, pairing always-on clean power with plans to extract lithium from geothermal fluids—though drilling costs remain a key constraint. Ingestible electronics: smart pill tech - Researchers are advancing swallowable “smart pills” that can sense biomarkers, transmit data wirelessly, and potentially deliver drugs or take biopsies, but power, miniaturization, and safety validation remain major hurdles. Streaming M&A: Netflix steps back - Netflix is walking away from a proposed purchase of Warner Bros. Discovery’s studio and streaming assets after the WBD board favored a higher all-cash bid from Paramount Skydance—reshaping the media consolidation map. Iran’s reported cruise missile talks - Reuters reports Iran is close to buying China’s CM-302 supersonic anti-ship cruise missile, a move unfolding alongside increased U.S. naval presence and renewed regional security pressure. Google and Adobe creative AI - Google is rolling out Gemini 3.1 Flash Image as its unified image generator, while Adobe’s Firefly video editor adds Quick Cut to auto-assemble rough drafts—pushing creative workflows toward faster iteration.
Episode Transcript
Pentagon pressures Anthropic on AI
First up: the standoff between the U.S. Department of Defense and Anthropic is escalating fast. Reporting from the AP and public statements from Anthropic CEO Dario Amodei paint the same basic picture—Defense Secretary Pete Hegseth wants Anthropic’s Claude models available for “all lawful” military uses, without carve-outs. Anthropic says it can’t agree “in good conscience” unless there are explicit safeguards against two specific categories: fully autonomous weapons and mass domestic surveillance.
What makes this different from the usual contract dispute is the language around leverage. Officials have floated labeling Anthropic a “supply chain risk,” and earlier threats even name-checked the Defense Production Act—the Cold War-era law that lets the U.S. government direct private industry for national defense. Legal experts are already signaling that using the DPA to override an AI company’s safety restrictions would be novel territory, and likely end up in court. The Pentagon says it’s not interested in illegal domestic surveillance or weapons without a human in the loop, but it also insists no vendor gets to dictate operational decision-making. The practical consequence: Anthropic could be offboarded, and rival labs that already accepted “all lawful purposes” terms could fill the gap.
Open source in age of cloning
Zooming out from government pressure to a different kind of pressure: the economic foundations of open source. Ghost founder John O’Nolan argues early 2026 is a weird, conflicted moment to maintain open source. On one hand, AI tools can reduce the grind—triage, refactors, docs, tests. On the other, AI is making it easier for competitors to clone what you’ve built.
O’Nolan points to a familiar open source story: your code is public, it gains traction, and eventually someone repackages it into a competing product—he cites Substack’s past reuse of portions of Ghost’s code. Maintainers respond with “follow our rules” strategies: tighter licenses, or keeping pieces private like test suites. But he says that strategy assumes code is scarce and expensive to produce. The uncomfortable twist is that frontier models may be breaking that assumption. He highlights Cloudflare’s claim that a single engineer rebuilt a Next.js-like system on Vite in roughly a week—raising the question: if a model can recreate the functionality without copying the literal code, what does a license even protect?
And his warning isn’t limited to open source. If your proprietary product has a public interface, O’Nolan expects AI-assisted reverse engineering to get good enough that cloning becomes routine—fast, cheap, and strategically weaponized.
AI science, math, and labs
Now to the other side of the AI story: not cloning, but discovery. OpenAI’s VP of Science, Kevin Weil, says we’re watching a familiar capability pattern play out again—things go from “impossible,” to working 5 or 10% of the time, to being reliably useful within six to twelve months. In his view, that arc is now reaching frontier science.
His most concrete vision is a closed loop between AI models and robotic labs: models propose experiments, simulations narrow the options, robots run protocols around the clock, results get ingested, and the next round improves via reinforcement-style iteration. The interesting point is not that this could happen, but that pieces of it already exist—robotic labs are real, and AI is already notching credible wins.
We’re seeing that in mathematics too. Researchers have used generative models to solve some previously unanswered problems, including items from Paul Erdős’ long list of questions. Terence Tao has reportedly served as an adjudicator for some AI-generated proofs, which gives the work weight—but he’s also trying to keep expectations grounded. Tao calls many of these results “cheap wins,” arguing the models are mostly picking off easier long-tail problems using known techniques. Still, even “cheap wins” add up, and Tao’s more optimistic note is that hybrid work—humans steering concepts while AI handles exhaustive exploration—could become the standard pattern.
AI competition, adoption, jobs
So is AI about to steamroll jobs and markets? Two separate takes this week pushed back on the simplest narratives.
Economist Jason Furman argues that, surprisingly, frontier AI looks more competitive than consolidated. Users can “multihome,” switching between models easily, and services increasingly route queries to the best model for the task. Furman points to sharp price declines—one estimate suggests GPT-4-level performance has fallen from around $20 per million tokens in 2022 to well under a dollar today. His caution is that hyper-competition can mean firms burning money for share, with margins squeezed as models and features commoditize.
Meanwhile, Citadel Securities’ Frank Flight says the data still don’t show an imminent wave of white-collar destruction. He emphasizes a key distinction: AI may improve recursively, but adoption doesn’t. Real-world diffusion tends to follow an S-curve because integration is expensive, workflows are sticky, regulations matter, and compute isn’t free. He points to stable measures of AI usage intensity at work, and even notes software-engineering job postings up year over year. The claim isn’t that displacement won’t happen—it’s that the timeline is being oversold, and the bottlenecks are practical as much as technical.
That S-curve theme also shows up in a widely shared explainer by Constance Crozier: fitting an S-curve early is notoriously unreliable. During the phase that looks exponential, the eventual plateau can be almost impossible to estimate without domain constraints. Translation: confident early forecasts—whether optimistic or apocalyptic—are usually fragile.
Chip wars: TPUs, High-NA
All of that feeds directly into the compute race—because even moderate adoption at scale demands an enormous amount of infrastructure.
On the dealmaking front, Meta is reportedly signing a multi-billion-dollar, multi-year arrangement to rent Google’s AI chips—TPUs—to train and build new models. It’s another signal that the market is expanding beyond Nvidia, with Meta also buying from AMD and continuing agreements with Nvidia. For Google, selling TPUs is a way to turn AI investment into cloud revenue, and it also pressures the broader accelerator ecosystem to become more interoperable.
On the geopolitics of optimization, Reuters reports DeepSeek—still famous for shaking markets with low-cost performance—has withheld pre-release access to its upcoming flagship model from Nvidia and AMD, while giving Chinese suppliers like Huawei an early tuning window. It’s a small operational move that hints at a bigger story: software stacks, compiler toolchains, and model releases are now part of industrial strategy.
And on the manufacturing side, ASML says its next-generation High-NA EUV lithography tools are ready for high-volume production use. These machines are staggering—about $400 million each—but the payoff is simplifying chip fabrication by replacing multiple steps with one. ASML says the tools have processed around half a million wafers and are hitting production-grade precision, though customers will still need years to integrate them into full manufacturing lines. If AI demand keeps rising, High-NA is one of the few credible paths to keep advancing chip density without hitting a wall.
What AI coding tools recommend
Let’s talk about what AI is doing inside software teams right now—because the “agent era” is arriving in very uneven, very opinionated ways.
Researchers at Amplifying.ai ran Anthropic’s Claude Code thousands of times across real repositories with open-ended prompts, then analyzed which tools it recommends. The headline: Claude often prefers to build rather than buy. “Custom/DIY” was the most common choice across categories—things like rolling feature flags with configs and rollout percentages, or building authentication in Python with JWTs and bcrypt-style libraries instead of selecting a managed platform.
But when it does pick a third-party tool, it’s decisive. Think GitHub Actions for CI/CD and Stripe for payments. The researchers also argue the model quietly standardizes a “default stack” that leans heavily JavaScript, and that newer model versions show a “recency gradient”—shifting recommendations toward newer frameworks and libraries.
That leads neatly into a framework from Kenneth Auchenberg: split agents into “Leverage” and “Functions.” Leverage agents make a human dramatically more productive—like an always-on chief of staff. Function agents aim to do an entire job end-to-end, with minimal human involvement. His point is that if you don’t choose which you are, your product strategy gets messy: pricing, buyer, trust model, and go-to-market all drift. Many products, he argues, start as leverage and only later “earn the right” to become function—copilot first, autopilot later.
Agents: copilot versus autopilot
Creative tools are also leaning into that “get me started” philosophy.
Google just rolled out a new image generation model—Gemini 3.1 Flash Image—which it’s positioning as both faster and near “pro” quality, with improved text rendering in images and better consistency across multiple characters and objects.
And Adobe is shipping a beta feature called Quick Cut inside its Firefly video editor. The idea is simple: give it your footage and a prompt, and it assembles a rough first draft—something editors can react to instead of staring at an empty timeline. Adobe is careful to frame it as a jumpstart, not a finished cut, but it’s another sign that “first pass generation” is becoming a standard interface pattern.
Tesla Cybercab leadership turnover
A quick set of non-AI headlines—still tech, still important.
At Tesla, Victor Nechita—the vehicle program manager for the Cybercab—says he’s leaving after nearly six years, right after Tesla marked the first production Cybercab coming off the line at Gigafactory Texas. The Cybercab is central to Tesla’s robotaxi ambitions, reportedly designed around autonomy from the start, potentially with no steering wheel or pedals, and with induction charging clearing a key regulatory step. Losing a program lead during a production ramp isn’t ideal, but it may also signal the project is shifting from heavy development into validation and scaling.
In Los Angeles, a bellwether trial over youth social media harm is underway. A 20-year-old lead plaintiff testified that she began using YouTube at six and Instagram at nine, describing anxiety, depression, self-harm, and body image issues worsened by filters and engagement mechanics. Plaintiffs argue features like infinite scroll and autoplay were intentionally addictive; Meta and YouTube deny wrongdoing and dispute causality. This case is the first of many trials that could shape how courts—and eventually regulators—treat algorithmic design and youth safety.
Social media addiction lawsuit trial
Two more science-and-energy stories worth your time.
In the UK, the country’s first geothermal electricity plant is switching on at United Downs in Cornwall. It taps deep granite heat from about five kilometers down, selling power into the grid—plus it plans to extract lithium from the geothermal fluids. It’s a compelling combo: always-on clean power and domestic battery materials. The open question is cost—deep drilling is expensive, so scaling beyond a few sites may be tough.
And in biomedical engineering, researchers are pushing “ingestible electronics”—capsules you swallow that can sense biomarkers, transmit data wirelessly, and potentially deliver drugs at a precise location or even take a tiny biopsy. The promise is less invasive diagnosis for GI disease, but the constraints are real: power budgets, reliable packaging against stomach acid, and proving safety and clinical value. Still, the direction is clear: diagnostics are trying to move from hospital procedures to at-home, single-pass monitoring.
Geothermal power and lithium extraction
Finally, two geopolitical and business items.
On the media side, Netflix is walking away from a deal to buy Warner Bros. Discovery’s studio and streaming assets after WBD’s board favored a higher all-cash bid from Paramount Skydance to acquire the entire company. Netflix says it stayed disciplined on price; the result is a consolidation fight that’s now less about content libraries alone and more about full-stack media empires—studios, streaming, and legacy networks.
And via Reuters, Iran is reportedly close to purchasing China’s CM-302 supersonic anti-ship cruise missile, designed to threaten large warships. The report lands amid increased U.S. naval presence in the region. It’s a reminder that “tech news” includes hardware in the most literal sense—and that advanced guidance and propulsion systems are part of today’s strategic calculus.
Subscribe to edition specific feeds:
- Space news
* Apple Podcast English
* Spotify English
* RSS English Spanish French
- Top news
* Apple Podcast English Spanish French
* Spotify English Spanish French
* RSS English Spanish French
- Tech news
* Apple Podcast English Spanish French
* Spotify English Spanish Spanish
* RSS English Spanish French
- Hacker news
* Apple Podcast English Spanish French
* Spotify English Spanish French
* RSS English Spanish French
- AI news
* Apple Podcast English Spanish French
* Spotify English Spanish French
* RSS English Spanish French
Visit our website at https://theautomateddaily.com/
Send feedback to [email protected]
Youtube
LinkedIn
X (Twitter)
- Effortless AI design for presentations, websites, and more with Gamma - https://try.gamma.app/tad
- KrispCall: Agentic Cloud Telephony - https://try.krispcall.com/tad
- Invest Like the Pros with StockMVP - https://www.stock-mvp.com/?via=ron
Support The Automated Daily directly:
Buy me a coffee: https://buymeacoffee.com/theautomateddaily
Today's topics: Pentagon pressures Anthropic on AI - The U.S. Defense Department is pressuring Anthropic to allow Claude for “all lawful” military uses, with threats including contract termination and “supply chain risk” labeling—raising major AI governance and safety-limit questions. Open source in age of cloning - Ghost founder John O’Nolan argues AI is undermining open source’s assumptions by making rapid re-implementation and reverse engineering easier, blurring whether licensing still protects maintainers when models can rebuild systems “from scratch.” AI science, math, and labs - OpenAI’s Kevin Weil described a near-term shift toward closed-loop discovery—models designing experiments, robotic labs running them 24/7, and results feeding back via reinforcement-like cycles—while new AI-assisted math proofs keep piling up under scrutiny from experts like Terence Tao. AI competition, adoption, jobs - Jason Furman says frontier AI is unusually competitive with easy model switching and rapid price deflation, while Citadel Securities’ Frank Flight argues adoption is still S-curve-shaped—so mass white-collar displacement isn’t showing up in labor data yet. Chip wars: TPUs, High-NA - Meta is reportedly renting Google TPUs in a multi-year deal as alternatives to Nvidia expand, DeepSeek is said to be optimizing first for Chinese chip suppliers, and ASML says High-NA EUV tools are ready for high-volume manufacturing—each reshaping compute supply. What AI coding tools recommend - A large benchmark of Claude Code’s tool choices suggests it often “builds, not buys,” frequently selecting Custom/DIY solutions while still strongly defaulting to certain third-party picks like GitHub Actions and Stripe—quietly standardizing stacks. Agents: copilot versus autopilot - Kenneth Auchenberg frames AI agents as either “Leverage” tools that multiply a person’s output or “Function” tools that replace a job end-to-end, a distinction that affects pricing, go-to-market, and the path from copilot to autopilot. Tesla Cybercab leadership turnover - Tesla’s Cybercab program manager Victor Nechita is departing during the ramp toward volume production, highlighting continued leadership churn as Tesla pivots harder toward autonomy, robotaxis, and AI-driven strategy. Social media addiction lawsuit trial - A bellwether trial in Los Angeles features testimony from a young woman who says social media addiction began in childhood, as plaintiffs argue features like infinite scroll and autoplay harmed youth mental health—claims Meta and YouTube deny. Geothermal power and lithium extraction - The UK switched on its first deep geothermal electricity plant at United Downs in Cornwall, pairing always-on clean power with plans to extract lithium from geothermal fluids—though drilling costs remain a key constraint. Ingestible electronics: smart pill tech - Researchers are advancing swallowable “smart pills” that can sense biomarkers, transmit data wirelessly, and potentially deliver drugs or take biopsies, but power, miniaturization, and safety validation remain major hurdles. Streaming M&A: Netflix steps back - Netflix is walking away from a proposed purchase of Warner Bros. Discovery’s studio and streaming assets after the WBD board favored a higher all-cash bid from Paramount Skydance—reshaping the media consolidation map. Iran’s reported cruise missile talks - Reuters reports Iran is close to buying China’s CM-302 supersonic anti-ship cruise missile, a move unfolding alongside increased U.S. naval presence and renewed regional security pressure. Google and Adobe creative AI - Google is rolling out Gemini 3.1 Flash Image as its unified image generator, while Adobe’s Firefly video editor adds Quick Cut to auto-assemble rough drafts—pushing creative workflows toward faster iteration.
Episode Transcript
Pentagon pressures Anthropic on AI
First up: the standoff between the U.S. Department of Defense and Anthropic is escalating fast. Reporting from the AP and public statements from Anthropic CEO Dario Amodei paint the same basic picture—Defense Secretary Pete Hegseth wants Anthropic’s Claude models available for “all lawful” military uses, without carve-outs. Anthropic says it can’t agree “in good conscience” unless there are explicit safeguards against two specific categories: fully autonomous weapons and mass domestic surveillance.
What makes this different from the usual contract dispute is the language around leverage. Officials have floated labeling Anthropic a “supply chain risk,” and earlier threats even name-checked the Defense Production Act—the Cold War-era law that lets the U.S. government direct private industry for national defense. Legal experts are already signaling that using the DPA to override an AI company’s safety restrictions would be novel territory, and likely end up in court. The Pentagon says it’s not interested in illegal domestic surveillance or weapons without a human in the loop, but it also insists no vendor gets to dictate operational decision-making. The practical consequence: Anthropic could be offboarded, and rival labs that already accepted “all lawful purposes” terms could fill the gap.
Open source in age of cloning
Zooming out from government pressure to a different kind of pressure: the economic foundations of open source. Ghost founder John O’Nolan argues early 2026 is a weird, conflicted moment to maintain open source. On one hand, AI tools can reduce the grind—triage, refactors, docs, tests. On the other, AI is making it easier for competitors to clone what you’ve built.
O’Nolan points to a familiar open source story: your code is public, it gains traction, and eventually someone repackages it into a competing product—he cites Substack’s past reuse of portions of Ghost’s code. Maintainers respond with “follow our rules” strategies: tighter licenses, or keeping pieces private like test suites. But he says that strategy assumes code is scarce and expensive to produce. The uncomfortable twist is that frontier models may be breaking that assumption. He highlights Cloudflare’s claim that a single engineer rebuilt a Next.js-like system on Vite in roughly a week—raising the question: if a model can recreate the functionality without copying the literal code, what does a license even protect?
And his warning isn’t limited to open source. If your proprietary product has a public interface, O’Nolan expects AI-assisted reverse engineering to get good enough that cloning becomes routine—fast, cheap, and strategically weaponized.
AI science, math, and labs
Now to the other side of the AI story: not cloning, but discovery. OpenAI’s VP of Science, Kevin Weil, says we’re watching a familiar capability pattern play out again—things go from “impossible,” to working 5 or 10% of the time, to being reliably useful within six to twelve months. In his view, that arc is now reaching frontier science.
His most concrete vision is a closed loop between AI models and robotic labs: models propose experiments, simulations narrow the options, robots run protocols around the clock, results get ingested, and the next round improves via reinforcement-style iteration. The interesting point is not that this could happen, but that pieces of it already exist—robotic labs are real, and AI is already notching credible wins.
We’re seeing that in mathematics too. Researchers have used generative models to solve some previously unanswered problems, including items from Paul Erdős’ long list of questions. Terence Tao has reportedly served as an adjudicator for some AI-generated proofs, which gives the work weight—but he’s also trying to keep expectations grounded. Tao calls many of these results “cheap wins,” arguing the models are mostly picking off easier long-tail problems using known techniques. Still, even “cheap wins” add up, and Tao’s more optimistic note is that hybrid work—humans steering concepts while AI handles exhaustive exploration—could become the standard pattern.
AI competition, adoption, jobs
So is AI about to steamroll jobs and markets? Two separate takes this week pushed back on the simplest narratives.
Economist Jason Furman argues that, surprisingly, frontier AI looks more competitive than consolidated. Users can “multihome,” switching between models easily, and services increasingly route queries to the best model for the task. Furman points to sharp price declines—one estimate suggests GPT-4-level performance has fallen from around $20 per million tokens in 2022 to well under a dollar today. His caution is that hyper-competition can mean firms burning money for share, with margins squeezed as models and features commoditize.
Meanwhile, Citadel Securities’ Frank Flight says the data still don’t show an imminent wave of white-collar destruction. He emphasizes a key distinction: AI may improve recursively, but adoption doesn’t. Real-world diffusion tends to follow an S-curve because integration is expensive, workflows are sticky, regulations matter, and compute isn’t free. He points to stable measures of AI usage intensity at work, and even notes software-engineering job postings up year over year. The claim isn’t that displacement won’t happen—it’s that the timeline is being oversold, and the bottlenecks are practical as much as technical.
That S-curve theme also shows up in a widely shared explainer by Constance Crozier: fitting an S-curve early is notoriously unreliable. During the phase that looks exponential, the eventual plateau can be almost impossible to estimate without domain constraints. Translation: confident early forecasts—whether optimistic or apocalyptic—are usually fragile.
Chip wars: TPUs, High-NA
All of that feeds directly into the compute race—because even moderate adoption at scale demands an enormous amount of infrastructure.
On the dealmaking front, Meta is reportedly signing a multi-billion-dollar, multi-year arrangement to rent Google’s AI chips—TPUs—to train and build new models. It’s another signal that the market is expanding beyond Nvidia, with Meta also buying from AMD and continuing agreements with Nvidia. For Google, selling TPUs is a way to turn AI investment into cloud revenue, and it also pressures the broader accelerator ecosystem to become more interoperable.
On the geopolitics of optimization, Reuters reports DeepSeek—still famous for shaking markets with low-cost performance—has withheld pre-release access to its upcoming flagship model from Nvidia and AMD, while giving Chinese suppliers like Huawei an early tuning window. It’s a small operational move that hints at a bigger story: software stacks, compiler toolchains, and model releases are now part of industrial strategy.
And on the manufacturing side, ASML says its next-generation High-NA EUV lithography tools are ready for high-volume production use. These machines are staggering—about $400 million each—but the payoff is simplifying chip fabrication by replacing multiple steps with one. ASML says the tools have processed around half a million wafers and are hitting production-grade precision, though customers will still need years to integrate them into full manufacturing lines. If AI demand keeps rising, High-NA is one of the few credible paths to keep advancing chip density without hitting a wall.
What AI coding tools recommend
Let’s talk about what AI is doing inside software teams right now—because the “agent era” is arriving in very uneven, very opinionated ways.
Researchers at Amplifying.ai ran Anthropic’s Claude Code thousands of times across real repositories with open-ended prompts, then analyzed which tools it recommends. The headline: Claude often prefers to build rather than buy. “Custom/DIY” was the most common choice across categories—things like rolling feature flags with configs and rollout percentages, or building authentication in Python with JWTs and bcrypt-style libraries instead of selecting a managed platform.
But when it does pick a third-party tool, it’s decisive. Think GitHub Actions for CI/CD and Stripe for payments. The researchers also argue the model quietly standardizes a “default stack” that leans heavily JavaScript, and that newer model versions show a “recency gradient”—shifting recommendations toward newer frameworks and libraries.
That leads neatly into a framework from Kenneth Auchenberg: split agents into “Leverage” and “Functions.” Leverage agents make a human dramatically more productive—like an always-on chief of staff. Function agents aim to do an entire job end-to-end, with minimal human involvement. His point is that if you don’t choose which you are, your product strategy gets messy: pricing, buyer, trust model, and go-to-market all drift. Many products, he argues, start as leverage and only later “earn the right” to become function—copilot first, autopilot later.
Agents: copilot versus autopilot
Creative tools are also leaning into that “get me started” philosophy.
Google just rolled out a new image generation model—Gemini 3.1 Flash Image—which it’s positioning as both faster and near “pro” quality, with improved text rendering in images and better consistency across multiple characters and objects.
And Adobe is shipping a beta feature called Quick Cut inside its Firefly video editor. The idea is simple: give it your footage and a prompt, and it assembles a rough first draft—something editors can react to instead of staring at an empty timeline. Adobe is careful to frame it as a jumpstart, not a finished cut, but it’s another sign that “first pass generation” is becoming a standard interface pattern.
Tesla Cybercab leadership turnover
A quick set of non-AI headlines—still tech, still important.
At Tesla, Victor Nechita—the vehicle program manager for the Cybercab—says he’s leaving after nearly six years, right after Tesla marked the first production Cybercab coming off the line at Gigafactory Texas. The Cybercab is central to Tesla’s robotaxi ambitions, reportedly designed around autonomy from the start, potentially with no steering wheel or pedals, and with induction charging clearing a key regulatory step. Losing a program lead during a production ramp isn’t ideal, but it may also signal the project is shifting from heavy development into validation and scaling.
In Los Angeles, a bellwether trial over youth social media harm is underway. A 20-year-old lead plaintiff testified that she began using YouTube at six and Instagram at nine, describing anxiety, depression, self-harm, and body image issues worsened by filters and engagement mechanics. Plaintiffs argue features like infinite scroll and autoplay were intentionally addictive; Meta and YouTube deny wrongdoing and dispute causality. This case is the first of many trials that could shape how courts—and eventually regulators—treat algorithmic design and youth safety.
Social media addiction lawsuit trial
Two more science-and-energy stories worth your time.
In the UK, the country’s first geothermal electricity plant is switching on at United Downs in Cornwall. It taps deep granite heat from about five kilometers down, selling power into the grid—plus it plans to extract lithium from the geothermal fluids. It’s a compelling combo: always-on clean power and domestic battery materials. The open question is cost—deep drilling is expensive, so scaling beyond a few sites may be tough.
And in biomedical engineering, researchers are pushing “ingestible electronics”—capsules you swallow that can sense biomarkers, transmit data wirelessly, and potentially deliver drugs at a precise location or even take a tiny biopsy. The promise is less invasive diagnosis for GI disease, but the constraints are real: power budgets, reliable packaging against stomach acid, and proving safety and clinical value. Still, the direction is clear: diagnostics are trying to move from hospital procedures to at-home, single-pass monitoring.
Geothermal power and lithium extraction
Finally, two geopolitical and business items.
On the media side, Netflix is walking away from a deal to buy Warner Bros. Discovery’s studio and streaming assets after WBD’s board favored a higher all-cash bid from Paramount Skydance to acquire the entire company. Netflix says it stayed disciplined on price; the result is a consolidation fight that’s now less about content libraries alone and more about full-stack media empires—studios, streaming, and legacy networks.
And via Reuters, Iran is reportedly close to purchasing China’s CM-302 supersonic anti-ship cruise missile, designed to threaten large warships. The report lands amid increased U.S. naval presence in the region. It’s a reminder that “tech news” includes hardware in the most literal sense—and that advanced guidance and propulsion systems are part of today’s strategic calculus.
Subscribe to edition specific feeds:
- Space news
* Apple Podcast English
* Spotify English
* RSS English Spanish French
- Top news
* Apple Podcast English Spanish French
* Spotify English Spanish French
* RSS English Spanish French
- Tech news
* Apple Podcast English Spanish French
* Spotify English Spanish Spanish
* RSS English Spanish French
- Hacker news
* Apple Podcast English Spanish French
* Spotify English Spanish French
* RSS English Spanish French
- AI news
* Apple Podcast English Spanish French
* Spotify English Spanish French
* RSS English Spanish French
Visit our website at https://theautomateddaily.com/
Send feedback to [email protected]
Youtube
X (Twitter)