Tech News · March 5, 2026 · 9:26

Android opens up Play Store & AI agents reshape developer tools - Tech News (Mar 5, 2026)

Android drops Play Store fees, Anthropic vs Pentagon fallout, Broadcom’s AI chip surge, Evo 2 genome model, and the EU’s social media age debate.

Android opens up Play Store & AI agents reshape developer tools - Tech News (Mar 5, 2026)
0:009:26

Our Sponsors

Topics

  1. 01

    Android opens up Play Store

    — Google is cutting Play Store fees and loosening rules around alternative billing and third‑party app stores, signaling a major shift in Android monetization under regulatory pressure.
  2. 02

    AI agents reshape developer tools

    — A wave of “agent-first” tooling ideas is emerging: machine-readable CLIs, runtime schema introspection, and cleaner documentation formats like WordPress.org Markdown for reliable AI automation.
  3. 03

    Pentagon pressure on AI labs

    — Vox reports the Pentagon blacklisted Anthropic after it refused surveillance and autonomous-weapon terms, while OpenAI moved onto classified networks—raising governance and contract-language debates.
  4. 04

    Chip supply risks and demand

    — South Korea warned Middle East tensions could disrupt helium and other chipmaking inputs, as AI-driven demand stays intense; Broadcom also projected massive growth in custom AI silicon revenue.
  5. 05

    Evo 2 genome foundation model

    — Evo 2, an open-source genome language model trained on a massive multi-species DNA dataset, aims to improve variant interpretation and genome annotation, with explicit biosafety trade-offs.
  6. 06

    EU weighs social media age ban

    — The European Commission convened an expert panel to consider an EU-wide minimum age for social media, taking cues from Australia and escalating platform compliance pressure.
  7. 07

    Nuclear and sensor tech milestones

    — TerraPower won a key US NRC construction permit for its Natrium reactor, while Duke engineers set a speed record for a new ultrathin light sensor—both notable for future energy and imaging.
  8. 08

    AI culture meets deepfakes

    — Two new documentaries spotlight AI’s cultural tension: optimism versus risk, and how deepfakes can impersonate public figures—fueling ongoing debates about consent and misinformation.

Sources

Full Transcript

Google is about to make Android a lot less “one store, one tollbooth” — and it could change how apps get paid for, and where they get downloaded. Welcome to The Automated Daily, tech news edition. The podcast created by generative AI. I’m TrendTeller, and today is March 5th, 2026. Let’s get into what moved the tech world in the last day or so — and why it matters.

Android opens up Play Store

First up: Android is getting a big policy reset. Google says it’s ending the old default of taking a third of Play Store transactions. The new approach lowers the standard cut on in‑app purchases, gives some developers a path to an even smaller share, and takes a lighter bite out of subscriptions. The bigger story isn’t just the percentage. Google is also loosening rules that previously boxed developers into Google’s billing. Apps will be allowed to offer alternative billing options inside the app, or steer users to complete purchases on the web. That’s a clear contrast with Apple’s more limited openings, and it’s another sign that regulators and courts are now shaping app economics as much as product teams are. Alongside that, Google is building a “Registered App Stores” program. Third‑party stores that meet safety and quality requirements should get a smoother install flow, even as basic sideloading remains possible — though Google is hinting it may put more friction on sideloading later in 2026. The rollout starts in the EEA, the UK, and the US by the end of June, then expands over time. And yes, Epic is already lining up for the moment: it says Fortnite will return broadly to Google’s store as these policies land.

AI agents reshape developer tools

Staying with the theme of power shifting toward users and developers, there’s a parallel conversation happening in AI tooling: command-line software built for humans is starting to look clumsy for AI agents. One developer, Justin Poehnelt, argues that “agent DX” is basically a different design target. Humans like forgiving interfaces and helpful hints. Agents need predictable behavior, clean machine-readable input and output, and security measures that assume the input might be dangerous — even when it’s coming from your own automation. His practical advice is to stop forcing agents through overly simplified flags and instead allow raw JSON payloads straight to APIs, so nothing gets lost in translation. He also points to something that sounds mundane but is crucial: tools that can describe themselves at runtime, so agents don’t rely on stale documentation shoved into prompts. And he emphasizes safety rails like dry runs and output sanitization, because agent workflows can turn small mistakes into fast, repeated mistakes.

Pentagon pressure on AI labs

That “agent-first” idea isn’t just theory. Google’s Workspace developer community has released an open-source CLI called gws that aims to be a single gateway to common Workspace APIs. What’s notable is that it doesn’t hardcode a fixed set of commands; it can discover capabilities dynamically, and it’s built to return structured data rather than pretty terminal output. It also includes a mode designed to plug into agent ecosystems, so an AI assistant can call Workspace actions like tools. The catch: it’s explicitly not an official Google product, and it’s under active development — two details that matter a lot if you’re considering it for anything mission-critical.

Chip supply risks and demand

In the broader “make machines better readers” department, WordPress.org has added a clean Markdown output for most pages. You can request Markdown directly, and pages can advertise that alternative format. This is partly about AI — making official documentation easier for models and agents to ingest, so they’re less likely to learn from outdated blog posts or scraped copies. But it’s also just a quality-of-life upgrade for developers who want docs in terminals, editors, or automated pipelines.

Evo 2 genome foundation model

Now for a cautionary tale on how AI collides with open-source licensing. A dispute has flared around the Python library chardet. Maintainers released a new version after using an AI coding tool to rewrite the codebase, then switched the license to MIT, framing it as a complete rewrite. The original author argues that may not be a clean break from the past, because exposure to the earlier code — by humans or by the AI process — can still make the result a derivative work under copyleft rules. It’s a messy, emerging question: if AI-assisted “rewrites” become a common path to relicensing, that could weaken copyleft protections, and it could also leave companies unsure what they’re actually allowed to ship.

EU weighs social media age ban

Let’s zoom out to geopolitics and AI policy, where the temperature is rising. Vox reports the Pentagon blacklisted Anthropic after the company refused to relax two “red lines”: no mass domestic surveillance and no fully autonomous weapons. The piece describes the move as a form of pressure on a private AI vendor — and it landed at the same time OpenAI announced work to deploy models on the Pentagon’s classified network. Even if you treat this as normal government procurement drama, it raises a very current issue: contract wording. Terms like “lawful purposes” can sound reassuring, but critics argue they don’t necessarily protect against large-scale surveillance enabled by modern data markets. The story also notes a growing worker backlash, with calls for solidarity across AI companies.

Nuclear and sensor tech milestones

China, meanwhile, is making its own intentions explicit. A new five-year policy blueprint unveiled around the National People’s Congress puts artificial intelligence all over the page, pairing an “AI+” push with goals in areas like quantum computing and robotics. Officials are framing this as a productivity strategy for an aging population, and also as a resilience strategy as export controls and tech rivalry deepen. The plan also nods to hyperscale computing buildouts and support for open-source AI communities — an approach Beijing seems to see as both an accelerator and a differentiator.

AI culture meets deepfakes

All of this runs on hardware, and hardware runs on supply chains. South Korea is warning that an escalating US–Israel conflict with Iran could disrupt key materials used in semiconductor manufacturing — with helium singled out as a particularly sensitive input. Even if companies have short-term inventories, the warning is about how quickly a geopolitical shock can ripple into production planning. At the same time, demand signals are still flashing bright green. Broadcom reported strong results and, more strikingly, its CEO said he expects next year’s AI chip revenue to blow past a huge milestone. The significance here is that the AI boom isn’t just about buying more standard GPUs; it’s increasingly about custom silicon and the industrial capacity to deliver it consistently.

In research news, one of the most ambitious open-source biology projects in a while just got louder. Researchers released Evo 2, a “genome language model” trained on an enormous dataset spanning bacteria, archaea, and eukaryotes. What makes this interesting is the direction of travel: genome modeling is moving beyond small, simpler organisms toward the long-range complexity of eukaryotic DNA. Early results suggest the model can score the impact of mutations in biologically meaningful ways, including tricky areas like splicing and noncoding regions, and the team has published the model and the dataset with some biosafety-minded exclusions. The near-term value is interpretation — faster annotation and better variant triage. Generation and design are still harder, and the researchers basically admit that: reading biology appears to be ahead of writing it, at least for complex organisms.

Two more quick hits before we wrap. The European Commission is convening an expert group to explore whether the EU should set a bloc-wide minimum age for social media. Recommendations are expected by summer 2026, and Australia’s under‑16 rule is clearly the reference point. If Brussels goes this route, it won’t just be a policy headline — it will force platforms to prove they can enforce age rules at scale, which is technically and politically thorny. And in hard science: TerraPower has a US Nuclear Regulatory Commission construction permit for its first plant in Wyoming — the first such approval in years. It’s a meaningful step for next-generation nuclear projects, even though it’s not an operating license, and timelines still look long. Finally, Duke engineers built an ultrathin light detector that’s unbelievably fast for its class and can sense across a very wide range of wavelengths. If it scales to manufacturing, it could lead to smaller, quicker multispectral imaging — useful in areas like medical screening and remote sensing.

One last note on AI in culture: two new documentaries are leaning into the public’s split-screen view of AI — breakthrough tool on one side, societal risk on the other. One film explores the emotional tension directly, while another uses an AI-generated “Sam Altman” stand-in to make a point about how easy impersonation and manipulation can be. Whatever you think of the filmmaking choices, the takeaway is clear: deepfakes and synthetic media aren’t a niche concern anymore. They’re becoming a mainstream literacy problem — for voters, consumers, and anyone who relies on “seeing is believing.”

That’s the tech landscape for March 5th, 2026: app store rules loosening, agent-first tooling getting more practical, and geopolitics pushing directly on AI and chips. If you want to keep one thread in mind today, make it this: a lot of the next year in tech will be decided less by raw capability, and more by who gets to distribute, who gets to bill, and who’s allowed to build. Thanks for listening to The Automated Daily, tech news edition. I’m TrendTeller — talk to you next time.