Daily TEA – Your Website Now Gets Graded by Robots
Chrome’s AI-agent score, Tesla’s Megapod, enterprise ROI reality check, a $7.5M bot hack, and a protocol for agents to find agents
Hello, dear TEA-mates! Here is what you need to know today.
1. 🤖 Chrome Now Scores How AI-Agent-Friendly Your Website Is
Google added an “Agentic Browsing” category to Lighthouse (the built-in Chrome auditing tool, requires Chrome 150+) that grades how well a site supports AI agents instead of humans. Rather than a single 0 to 100 score, it returns pass/fail ratios across three areas: WebMCP tool registration (whether the site exposes its logic to agents via declarative HTML or JavaScript tools), agent-centric accessibility (programmatic names, valid ARIA roles, parent-child relationships, and content visible to the accessibility tree), and stability/discoverability (Cumulative Layout Shift plus the presence of an llms.txt file). Google frames the goal as gathering data and giving actionable signals rather than a definitive ranking. Its top recommendations: adopt WebMCP, prioritize semantic HTML and proper ARIA labeling (the “machine-eye view”), and reduce layout shifts so agents can interact reliably. (Read More)
🫖 TEA For Thought: “This is amazing. Adopt it to your website now to make it more AI-agent friendly.”
2. ⚡ Tesla Files “Megapod” Trademark to Sell Its AI Data Center Hardware
Tesla filed a trademark for “Megapod,” described as modular data center hardware systems for AI computing that bundle servers, networking, power distribution, and cooling into self-contained units for AI workloads. The filing signals Tesla wants to package and sell the data center expertise it built for its own AI efforts. The article notes Tesla’s real edge may sit in power rather than compute: xAI has bought roughly $1 billion of Tesla Megapack energy storage, and Tesla’s Cortex cluster runs about 67,000 Nvidia H100-equivalent GPUs. Context on the compute side is rougher: Tesla killed its Dojo supercomputer in August 2025, its AI5 chip taped out nearly two years late, and the AI6 chip slipped about six months, with mass production now targeted for late 2027. No pricing, capacity, or build-time figures were disclosed for Megapod itself. (Read More)
🫖 TEA For Thought: “Tesla’s 200 days to finish building AI data centers was already a miracle. It’s great to see they are trying to SaaS it and monetize on their hard work.”
3. 📉 Enterprises Are Going All-In on AI While Quietly Cutting Their ROI Timelines
TEKsystems’ new State of Digital Transformation 2026 survey shows companies deepening their AI commitment even as they get more sober about payback. Enterprise-wide AI adoption doubled to 24% from 12% a year earlier, yet the share of organizations expecting a return within six months fell to 27% in 2026 from 42% in 2025. Spending plans stayed aggressive: 71% of all organizations expect to raise AI budgets, rising to 76% among the digital leaders versus 61% of laggards (the laggards themselves up from 48% a year ago), and those leaders reported roughly 2.5 times more confidence than laggards that their investments would hit ROI targets. The report reads the cooling of fast-return hopes as a sign the easy wins have been claimed, leaving open whether wider adoption converts into real operational change or whether a longer ROI horizon simply becomes the new normal. (Read More)
🫖 TEA For Thought: “There’s no such thing as a quick win.”
4. 🥪 A $7.5M Ethereum Trading Bot Got Tricked Into Authorizing Its Own Robbery
The well-known Ethereum MEV bot operating under the address jaredfromsubway.eth, one of the network’s most active sandwich-trading operations since early 2023, lost roughly $7.5 million in a security incident on June 20, 2026. According to onchain security firm Blockaid, the attacker did not use phishing, a stolen private key, or a DeFi protocol bug. Instead, they manipulated the bot’s automated trading logic into authorizing attacker-controlled smart contracts, turning the bot’s own decision-making against it. The case highlights a distinct class of risk for autonomous trading systems: even sophisticated, long-running operators can be compromised through their automation rather than through traditional credential theft. (Read More)
🫖 TEA For Thought: “More like this will happen as AI gets more and more powerful.”
5. 🧭 A New Protocol Lets Agents Find Other Agents (and People Find Agents Too)
InfoWorld breaks down Agentic Resource Discovery (ARD), a proposed open protocol for standardizing how AI agents discover the tools and services they need inside enterprises. The problem it targets: when an agent investigates a production issue, it may need engineering docs, support tickets, deployment history, and observability data scattered across siloed systems with no unified discovery layer. ARD works on two levels: organizations publish catalogs describing their available capabilities, and registries act as search engines that crawl and index those catalogs. The specification is public, with a quickstart for publishing catalogs and an open-source community, and it is backed by Google, Microsoft, Cisco, Nvidia, and Salesforce. (Read More)
🫖 TEA For Thought: “This is essentially a protocol for sharing agents, for people to find agents and for agents to find agents. Not A2A but AfA.”
🛠️ Skill of the Day
The Pre-Mortem Planner: imagine your plan already failed, then work backwards to fix it before you start.
You are a seasoned project planner running a pre-mortem on my plan.
Here is what I am about to do:
[DESCRIBE YOUR PROJECT, DECISION, OR LAUNCH IN A FEW SENTENCES]
Do this in four steps:
1. Assume it is six months from now and this plan has clearly failed. Write a short, blunt story of HOW it failed, as if it already happened.
2. List the 7 most likely specific causes of that failure, ranked from most to least probable. Be concrete, not generic (no "poor communication").
3. For each cause, give one early warning sign I could notice in the first two weeks, and one cheap action I can take now to prevent or reduce it.
4. End with the single biggest risk I am probably underestimating, and one honest question I should answer before I commit.
Be direct. Do not reassure me or pad the answer. If the plan has a fatal flaw, say so plainly.
Paste into ChatGPT, Claude, or your tool of choice. Replace the bracketed bit with your own plan, and the more specific you are, the sharper the warnings.
TEAHEE Moment
Stay sharp, stay informed. See you tomorrow.
If you enjoyed this TEA, follow along on social for more:
Twitter/X