Please support this podcast by checking out our sponsors:

- Effortless AI design for presentations, websites, and more with Gamma - https://try.gamma.app/tad

- Discover the Future of AI Audio with ElevenLabs - https://try.elevenlabs.io/tad

- KrispCall: Agentic Cloud Telephony - https://try.krispcall.com/tad

Buy me a coffee: https://buymeacoffee.com/theautomateddaily

Today's topics:

Dictionary compression reshapes web delivery - HTTP dictionary compression via RFC 9842 and Zstandard/Brotli dictionaries can slash repeat-visit payloads—tests cite up to 90% smaller JavaScript than Brotli, with Chrome 130+ leading support.

Encrypted RCS finally crosses platforms - Apple and Google are testing end-to-end encrypted RCS between iPhone and Android under the GSMA Universal Profile, bringing lock-icon encryption to cross-platform chats in iOS 26.4 beta and Google Messages beta.

Meta’s massive AMD chip bet - Meta signed a five-year, $60B purchase agreement for AMD AI hardware and took a 10% stake, signaling a major shift toward inference-optimized accelerators like MI450 and multi-vendor compute strategy.

Model distillation, data, and security - Anthropic alleges Chinese startups used 24,000 fraudulent accounts to generate 16M Claude conversations for distillation, raising terms-of-service, safety-guardrail, and national security concerns around model copying.

AI coding era: quality and juniors - Microsoft leaders warn AI agents can boost seniors but slow juniors, while Simon Willison pushes red/green TDD as an agent prompt; the shared theme is verification, tests, and mentorship to avoid brittle code.

Robots, Mars autonomy, and navigation - China showcased more capable humanoid robots amid booming shipments, while NASA gave Perseverance a Mars ‘GPS-like’ localization upgrade—both point to faster autonomy, but with real-world robustness still a question.

Geothermal’s scale test and economics - Geothermal companies are diverging: EGS drilling gains may plateau while closed-loop systems target district heating economics; scalability hinges on rigs, capex, water losses, and siting near transmission or data centers.

Gene editing precision and FDA pathway - Researchers reduced base-editing bystander mutations by over 80% using linker redesigns, as the FDA detailed a ‘plausible mechanism pathway’ for bespoke ultra-rare genetic medicines—promising, but still bounded and early.

Export controls and tech geopolitics - China imposed export restrictions on Japanese entities tied to military capability, and Washington launched a Peace Corps-style Tech Corps to promote US AI abroad—highlighting the tightening tech-and-supply-chain contest.

IPO float shock and software selloff - A scenario analysis about an ‘intelligence crisis’ sparked a software-stock selloff via SaaS pricing risk, while a separate IPO-float argument says SpaceX/OpenAI/Anthropic public listings could strain index flows and market liquidity.

First up: dictionary compression is quietly becoming one of the most meaningful performance upgrades the web has seen in years—especially for returning users. The concept is simple: instead of compressing a response from scratch every time, the server and client share a pre-arranged “dictionary,” so the compressed payload can mostly point to bytes the browser already has.

What’s new is that this is now getting standardized and deployed in a way that actually works on today’s internet. Early real-world results are eye-catching: YouTube’s desktop JavaScript bundle for returning users has been reported to shrink by as much as 90% compared to Brotli, and Google Search results HTML nearly 50% smaller for returning visitors—about 23% smaller on average once you include first-time users and the cost of downloading the dictionary.

If this sounds familiar, it’s because dictionary-based ideas have been around since the 1990s. The web even tried a version of it with Google’s SDCH, which was ultimately dropped by 2017 amid security, privacy, and deployment headaches.

The difference now is modern tooling and modern standards. Zstandard was designed with dictionary support from the start, Brotli gained official dictionary support in 2023, and—crucially—the IETF’s Compression Dictionary Transport spec, RFC 9842, defines how browsers and servers negotiate dictionaries over HTTP. That includes headers like Available-Dictionary, Use-As-Dictionary, and the right caching behavior via Vary, so you don’t accidentally serve the wrong compressed bytes to the wrong user.

Where does it stand today? Chrome 130 and other Chromium-based browsers are the strongest beachhead. Safari and Firefox have publicly signaled intent, but the key point is: this can be shipped as an opt-in enhancement. If your browser doesn’t support it, you just get the normal Brotli or Zstandard response.

One practical pattern highlighted for JavaScript bundles is to treat older bundle versions as dictionaries: store them keyed by hash, advertise them as usable dictionaries, and when a client offers a match, send a dictionary-compressed response. The big caveats: same-origin restrictions exist for good privacy reasons, caching has to be done carefully, you pay some complexity on the server, and shrinking transfers doesn’t make JavaScript parse and execute any faster on the device. Still, for bandwidth and latency, this is a serious lever.

Staying on communications: Apple and Google say cross-platform end-to-end encryption for RCS is now in testing. This matters because RCS has already improved the iPhone-to-Android experience with typing indicators, read receipts, and better media—thanks to Apple adding RCS in iOS 18—but encryption was the missing piece.

In the current beta, iPhone users on iOS 26.4 beta 2 with supported carriers can see an “Encrypted” label and lock icon for RCS chats, and Android users in the Google Messages beta should see a similar lock indicator. It’s still beta software, so glitches are expected, but the direction is clear: encrypted messaging is becoming the baseline across the Android–iPhone divide, not a platform perk.

Now to AI infrastructure, where the checks keep getting larger. Meta has agreed to a five-year deal to buy sixty billion dollars’ worth of AI chips from AMD—and it’s also taking a 10% stake in the chipmaker. AMD says the order represents 6 gigawatts worth of chips delivered over time, starting with 1 gigawatt of the upcoming MI450 hardware in the second half of this year.

The interesting detail: MI450 is described as optimized for inference—the day-to-day generation step that powers chatbots and assistants. Analysts increasingly expect inference to become the bigger market than training. Meta is also buying from Nvidia, and reportedly exploring options like Google’s TPUs, which reinforces a theme for 2026: the biggest AI buyers don’t want a single point of failure in their compute supply.

On the model side, Anthropic says it has evidence that three Chinese AI startups—DeepSeek, Moonshot, and MiniMax—improperly harvested Claude output at scale. The claim is that around 24,000 fraudulent accounts generated more than 16 million conversations, potentially usable for distillation, which is the technique of training one model to mimic another.

Anthropic frames this as more than a business dispute: it argues distillation can strip away safety guardrails and could amplify national-security risks. The accused companies haven’t publicly responded, but the pattern itself is becoming familiar across the industry—labs are hardening rate limits, fraud detection, and policy enforcement, while governments get pulled into a debate that’s part IP, part safety, and part geopolitics.

Let’s talk about what AI is doing to software work itself. Microsoft’s Mark Russinovich and Scott Hanselman are warning about a “hollowing out” risk: AI agents can give senior engineers a boost, but create drag for early-career developers who have to validate, integrate, and clean up agent output. They argue the predictable business reaction—hiring fewer juniors—could starve the future pipeline of experienced technical leaders.

Related, Simon Willison has been pushing a very actionable tactic: when you prompt a coding agent, explicitly ask for red/green test-driven development. In other words, write tests first, confirm they fail, then implement until they pass. It’s a compact instruction that forces the agent to prove behavior rather than just generating plausible-looking code. Across both takes, the message is the same: code may be cheaper to produce, but verification—tests, reviews, mentoring—still costs real effort.

Meanwhile, investors are trying to price the second-order effects. A widely shared scenario analysis—framed as a stress test, not a prediction—sparked a sharp selloff in a range of software stocks. The hypothetical argues that by 2027, some SaaS vendors could face brutal renewal negotiations as customers swap tools for AI-first alternatives or build internal systems.

And in a separate market-structure argument, another piece looks ahead to potential mega-IPO waves from SpaceX, OpenAI, and Anthropic. The headline valuations are huge, but the claimed real bottleneck is float: public markets may not be able to absorb a typical 15 to 25 percent IPO float without major disruption. If companies list with smaller floats, they may not qualify for major indices at first—then create a new kind of shock later when index inclusion forces passive funds to buy in and sell other holdings to make room.

Switching gears to robotics and autonomy: China’s humanoid-robot scene is being portrayed as moving fast—from awkward demos to much more polished performances, including choreographed routines and backflips during the Spring Festival Gala. Commentators point to EV-era supply chains—motors, batteries, sensors—plus better AI control systems as accelerants.

The sober counterpoint is that stage choreography is not the same as robust, perception-rich work in messy industrial settings. There are also familiar concerns: privacy if these machines are packed with sensors, job displacement, and the risk of an investment bubble if expectations outpace real deployments.

And speaking of autonomy done the hard way: NASA has upgraded the Perseverance rover with something close to a Mars version of GPS—without any GPS satellites. The new capability, called Mars Global Localization, lets the rover match its own panoramic images to onboard orbital maps and compute its position on its own.

NASA says it can localize to within about 10 inches, and the algorithm takes roughly two minutes to run. That matters because navigation drift can force the rover to stop early near hazards while Earth-based teams confirm its location—a process that can cost a whole Martian day. More autonomy here means more daily driving and more science, and it builds on NASA’s recent experiments with AI-planned routes.

On climate-tech and energy: a new geothermal sector update argues the field is entering a phase where company strategies start to diverge. In enhanced geothermal systems—EGS—drilling performance is still improving, but some gains may plateau as the easiest time savings get captured. There’s also a push toward hotter, deeper reservoirs to improve output, with practical constraints like high-temperature-rated components and water losses.

Closed-loop geothermal—the “underground radiator” approach—gets a more optimistic note for district heating, especially in Europe. One project in Germany is now operational providing combined heat and power, but electricity economics remain tougher, largely because of power-plant cost and low conversion efficiency. The through-line for both approaches: the next few years are the “great filter,” where scalability and project economics will decide whether this becomes a meaningful slice of generation or stays niche.

Finally, in biotech policy and research, there are two developments that connect. Researchers at Penn and Rice refined a base-editing technique to cut unintended “bystander” DNA edits by more than 80% while keeping strong on-target activity. In cystic-fibrosis-relevant sites, they report reducing bystander edits from as high as 50 to 60 percent down to below 1 percent in lab tests—still early, but it’s the sort of precision improvement gene-editing needs to move safely.

At the same time, the FDA released detailed guidance for a “plausible mechanism pathway” aimed at approving bespoke medicines for patients with ultra-rare, sometimes unique mutations—cases too small for traditional trials. It’s a big policy signal: the agency is trying to turn one-off genetic success stories into a repeatable pipeline, while still putting boundaries on how that pathway can be used.

One quick geopolitics note to close: China’s Commerce Ministry has published new export-control lists targeting Japanese entities tied, in China’s view, to military capability. Separately, the White House launched a Peace Corps-style “Tech Corps” initiative intended to deploy tech-skilled volunteers abroad to help implement American AI solutions. Put together, it’s another reminder that in 2026, chips, software, standards, and talent are all being treated as strategic assets—not just commercial ones.

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)