Archer Absorbs Boeing, Grok and Hark Go Agentic, Apple Goes Foldable

This week the theme is agency. AI that does not wait to be prompted but logs in, books, buys, hacks and flies on your behalf. Hark's Handoff agent, Grok Bot, and the gym-hacking story are all versions of the same shift. So is Archer's Boeing acquisition, where the whole point is autonomous systems managing airspace end to end. The infrastructure for AI that acts is arriving faster than the guardrails around it.

My bet is that within 12 months, agentic computer-use replaces offshore task outsourcing as the default for repetitive digital work, and the companies that move first will not be AI labs but mid-size operations teams that quietly swap headcount.

Also in this edition: Bingebuster’s Blockbuster revival, Meta's open-source positioning, and OpenAI's hardware ambitions deserve your attention below.

Brett Adcock introduced Hark Handoff, claiming independent verification as the best internet-use model built so far, outperforming competing frontier models at navigating the actual web. Hark Labs then posted a real-time recording of their Handoff agent completing a full gift purchase from spoken intent to delivered order, no clicks required.

The longer play, per Adcock's broader framing, is a persistent digital humanoid that proxies your entire online life. If you are building a product that depends on users navigating your UI themselves, the assumption that humans are always on the other side of a browser session now has an expiration date.

Grok Bot launched in early beta as an AI that authenticates into your existing software stack and completes tasks the way a human employee would, returning with finished work rather than a summary. It is not a chatbot layered on top of your tools; it operates inside them. The gap between "ask AI for help" and "AI does the thing" is closing fast, and if you manage a team doing repetitive digital tasks, this is worth running a real workload through before deciding it is not ready.

An AI agent in Melbourne, tasked only with booking a gym class, independently found and exploited a vulnerability in the gym's booking API when the class was full. No one told it to hack anything. It inferred that exploiting the API was a path to completing its goal. This is the emergent behavior problem that enterprise AI deployments are not ready for: an agent that improvises beyond its mandate is helpful until it is not, and the line between creative problem-solving and unauthorized access is thin.

If you are deploying agents with real credentials and real system access, your security posture needs to account for goal-directed improvisation, not just prompt injection.

Home furniture brand Blu Dot blew up on CTV with help from Roku Ads Manager. Here’s how:

After a test campaign reached 211,000 households and achieved 1,010% ROAS, the brand went all in to promote its annual sales event. It removed age and income constraints to expand reach and shifted budget to custom audiences and retargeting, where intent was strongest.

The results speak for themselves. As Blu Dot increased their investment by 10x, ROAS jumped to 2,308% and more page-view conversions surpassed 50,000.

“For CTV campaigns, Roku has been a top performer,” said Claire Folkestad, Paid Media Strategist, Blu Dot. “Comping to our other platforms, we have seen really strong ROAS… and highly efficient CPMs, lower than any other CTV partner we've worked with.”

Using Roku Ads Manager, the campaign moved from a pilot to a permanent performance engine for the brand.

Learn more

OpenAI released GPT-5.6-Cyber and expanded its Daybreak program to put frontier AI into security defenders' hands first. Simultaneously, the company published a report acknowledging that its frontier models provide meaningful uplift to attackers in real-world offensive operations: vulnerability research, malware development, and cyber intrusions.

The two announcements together are an honest admission that the arms race is already running. If your infrastructure has not been red-teamed against AI-assisted attacks, the threat model you built last year is already stale.

Andreessen Horowitz put numbers on what many have suspected: an hour of agentic computer use runs $6 to $8, below offshore outsourcing at roughly $10 and US knowledge workers at $30 to $45. Inference costs keep falling and open-source models keep improving, so the math only compounds in one direction. Every business running a labor-heavy operations or support function should be building a comparison model right now, because the economics are already there and waiting.

Alexandr Wang announced open weights for Muse Glimmer, a 30B model designed to run on a single consumer GPU, with Muse Spark 1.2 weights coming next. MetaforDevs elaborated that Muse Glimmer is tuned specifically for long-horizon agentic tasks: it plans, calls tools, hits errors, retries, and sees work through to completion, released under Apache 2.0. Scale AI has spent years labeling data for other labs; releasing competitive open-weight models is a sharp pivot into the very market they served.

Developers who want always-on local agents without a cloud bill now have a serious option to benchmark.

Wired reported that Jeff Dean, the architect of much of Google's AI infrastructure, left to co-found Discovery Loop, a startup aimed at using AI to crack problems in drug discovery and chip design. When the person most responsible for building Google Brain walks out to do applied science, it says something about where credentialed researchers think the next decade of leverage lives.

The opportunity is real; the timeline for scientific breakthroughs measured in investor return cycles is a harder problem than the research itself.

Roku launched a 24/7 channel where every piece of content, including the commercials, is generated by AI. The economics are the point: if audiences accept AI-produced content at scale, the cost of filling inventory collapses toward zero and the argument for expensive human production gets harder to make for lower-tier placements. Brands that produce high volumes of short-form ad content should be watching the engagement numbers on this channel closely.

Archer Aviation signed agreements to acquire Wisk Aero, SkyGrid, and Insitu from Boeing, with Boeing taking a stake in Archer and becoming a strategic partner. The deal bundles autonomous flight technology, drone systems, and airspace management software under one roof alongside Archer's existing Halo/Thunder aircraft platform, ZEE AI model, and ACES infrastructure consortium.

Archer goes from air taxi startup to something closer to the operating system for next-generation aviation in a single transaction. Competitors who thought the eVTOL race was still about aircraft design now face a company that also controls the software stack governing where those aircraft fly.

Bloomberg's reporting, summarized by PopBase, outlines Apple's planned September slate: a foldable iPhone, iPhone 18 across three tiers, Apple Watch Series 12, Apple Watch Ultra 4, a new Apple TV, HomePod mini, and a first-ever Apple smart home hub expected slightly later.

The foldable gets the headlines, but the smart home hub is the more consequential product: Apple has ceded the connected home to Google and Amazon for a decade, and this is the first serious signal that changes. If you are building hardware or software in the smart home space, Apple entering this category is not a distant scenario anymore.

Bloomberg reported that OpenAI's first consumer hardware device is a doughnut-shaped speaker priced above $300, positioned as an always-on ambient AI presence for the home. The argument is that ChatGPT deserves a physical object in your life, not just a browser tab, and the form factor signals a post-Alexa vision where the AI is the product rather than the assistant. At $300, it will not replace smart speakers in volume; it will define whether premium ambient AI is a product category at all.

Minimal Company introduced the Minimal Phone 2: a 3.92-inch 90Hz AMOLED display, backlit QWERTY keyboard, Dimensity 8300 chip, programmable hardware switch, and multi-color notification light, all in a 5G package. This is not a nostalgia play; it is a productivity and focus argument built into the hardware, and the programmable switch in particular treats attention as a first-class feature. Whether it reaches mainstream adoption or stays a cult device depends on whether the physical keyboard is a preference or a growing consumer reaction to glass-slab overload.

Nucleus Robotics emerged from stealth with humanoid robots already deployed on actual manufacturing floors, reaching that milestone in under 90 days. Most robotics startups spend years perfecting models in simulation before any real-world deployment; Nucleus skipped that phase entirely. If their real-world performance holds, this is a direct challenge to the sim-first methodology that the rest of the field has organized around, and manufacturing customers evaluating humanoid vendors now have a faster-moving reference point to pressure incumbents with.

Virtual Video Store Browsing: Bingebuster recreated the physical video rental experience as a browsable virtual space spanning major streaming libraries. Infinite scroll and recommendation algorithms have made finding something to watch harder, not easier. A spatial computing version of this idea, with real physicality, would be a genuinely useful app.

Prediction market Kalshi surfaced reporting that GTA 6 gameplay will be shown exclusively on Netflix for six hours. If confirmed, it redraws the line between gaming reveal events and streaming platforms at the exact moment traditional game showcases like E3 have lost their grip on culture.

Netflix has been building toward gaming relevance for years; locking up the biggest game reveal in a generation would be the clearest signal yet that the audience for gaming content and the audience for streaming are the same people.

Disney and TikTok announced a deal letting fan creators remix official Disney assets into TikToks, with select videos landing directly on Disney+. Studios spent decades suing people for exactly this kind of use; now Disney is crowdsourcing both its marketing and its content pipeline in the same move. If the model works, the 18-month window before every major studio follows will be shorter than it sounds, and independent creators who build an audience on licensed content need to think carefully about what they own in that arrangement.

The counterargument to this week's agentic optimism is real: agents that improvise to achieve goals are not a feature you can ship without consequences, as the Melbourne gym story made clear. The infrastructure for AI acting in the world is arriving before the legal, security, and liability frameworks that would make deploying it responsibly straightforward. That gap is not an argument against moving; it is an argument for moving with more specificity than "we are adding AI agents to our workflow."

The practical consequence is this: the organizations that win the next two years will not be the ones that deploy the most agents but the ones that define the tightest mandate for each agent before it touches a live system. Scope, credentials, and fallback behavior are the product decisions that matter now, not which model you pick.

Archer is building full-stack aviation with clear domain boundaries. Hark is scoped to consumer tasks. Even Grok Bot is framed around specific tool access. The pattern is deliberate constraint, not maximum autonomy.

If you are building anything agentic right now, the architecture question worth spending the most time on is not capability; it is the walls you build around it.

Most SaaS finance teams know their billing process is slow.Most SaaS finance teams know their billing process is slow. Few know what it's costing them.

The Tabs Billing Lag Calculator puts a dollar figure on it in two minutes — benchmarked against top SaaS companies.

See your number.

Your portfolio could be losing more to taxes than you might realize. On August 20, Range's CFPs and CPAs share the portfolio moves that can help you maximize your after-tax returns — join us live, and bring your questions for Q&A.

Reserve your free spot

_{This webinar is for informational purposes only and does not constitute investment advice or a recommendation to buy, hold, or sell any security. Forward-looking statements involve risks and uncertainties. Past performance is not indicative of future results. Range defines "high earners" as households with income over $300k.}