Econify.AI Newsletter, November 6, 2026

Musk calls Grokbot's costs unsustainable

Grokbot, the personal agent (an AI assistant that takes actions on your behalf) from Elon Musk's AI business, started hitting usage limits over the weekend. Musk responded that the cost of serving it was unsustainable relative to revenue. Skeptics read that as proof AI vendors have priced plans below cost, with investors covering the gap. One analysis cited this summer put a month on a $200 ChatGPT Pro plan at about $14,000 at regular pay-per-use prices.

Econify's take: Flat-rate plans are the exposed part of your AI budget. Several companies, including Apple, are putting their organizations on a budget. Start thinking about which AI tasks are actually important and what they are worth to you.

Voters turned against local data centers

Democrats won the midterms, and the effect on AI buyers runs through data centers. Coverage of Pew polling found 61% of voters dislike candidates who support building them locally. Voters sided against data centers in 38 of the 41 local ballot measures. One, in Harwood, North Dakota, went the industry's way, and the rest were uncounted as of Wednesday. Nine of the 10 largest states now have Democratic governors. Texas's governor-elect campaigned on a moratorium until rules are written, after Greg Abbott paused new permits in September. Commentators expect no federal moratorium to pass, because President Trump said that if Democrats won, he would veto. More likely are rules on incident reporting, whistleblower protection, vendor liability, and codified pre-release testing of frontier models.

Econify's take: We'd bet the near-term effect is tighter, pricier capacity in a few states, with little change to which tools you can use this year. We wouldn't rewrite AI plans yet, but be on the lookout for increases in incidents and outages.

An AI-run law firm is bidding for the files of a collapsed Wall Street firm

When Weil, a 95-year-old Wall Street firm, announced plans for an orderly wind-down this week, The Information reported that legal AI startup Crosby submitted a letter of intent to buy some of its assets, including its brand and internal data. Crosby also hopes to hire some of Weil's 1,200 lawyers. The New York Times attributed the collapse to partner departures, a failed succession plan, and a scuttled merger. Crosby, last valued at $400 million, operates as a law firm with about 50 lawyers and its own clients. Its bet is that precedent documents, contract templates, and workflows form a playbook worth real money. Ryan Walker, CEO of Gen Legal, disagreed, saying the data belongs to clients and can't be acquired.

Econify's take: This story is about institutional knowledge and training a specialized AI to use it. The templates and procedures your best people carry, and their past work, are what AI tools learn from, and most mid-market companies haven't written them down.

Frontier models can improve open source ones

Researchers reported that OpenAI's GPT-6 Astra can read the raw numbers that define an AI model (its weights) and rewrite a few directly, with no retraining. In the headline test, Astra changed a tiny fraction of the settings in Qwen3.5-0.8B, a very small open model, to stop it from getting stuck in reasoning loops. Its math score rose about sixfold and instruction following roughly tripled, as reported. A Google researcher reportedly replicated the result while noting these models are tiny next to frontier systems. Weijia Xu of Microsoft Research said the gain came from fixing a loop bug, which says little about general math ability. The authors add that the method couldn't improve a second small model that rarely looped.

Econify's take: In addition to self-improving AIs, let's also be on the lookout for AIs that can uplift less capable models. Wild times ahead if either of those things become mainstream.

Other news

Anthropic's amended IPO filing says it may soon meet its own threshold for AI-driven recursive self-improvement, and that it is considering policy updates to keep humans in control of frontier model development. Factory's Droid 1.0 makes programmatic tool calling the default, and Factory reports it cuts Droid's cost by 15% on Claude Opus 5.5. Spawn's founder described a ten-person company whose agent fleet reportedly shipped 77,000 code changes in a month on $3.1 million of token spend. Azure revenue reached $33.2 billion, yet Microsoft and Google both reported capex below analyst estimates, with chip delays cited, and Alphabet shares fell in extended trading. The Justice Department is asking about Nvidia's $13 billion Hugging Face purchase as part of a broader multiyear probe, and no charges have been filed. Binance suffered a record $6.3 billion hack, which TRM Labs links to North Korea's TraderTraitor group. To keep AI-assisted work from becoming workslop, the low-quality output BetterUp Labs and Stanford researchers named, practitioners suggest forming a view before opening the tool and reviewing at planned checkpoints.