🌀🗞 The FLUX Review, Ep. 231

April 16th, 2026

“FCP-230” // Photo: MK with Midjourney

Episode 231 — April 16th, 2026 — Available at read.fluxcollective.org/p/231

Contributors to this issue: Erika Rice Scherpelz, Ade Oshineye, Ben Mathes, Neel Mehta, MK, Boris Smus

Additional insights from: Dart Lindsley, Jasen Robillard, Jon Lebensold, Justin Quimby, Lisie Lillianfeld, Robinson Eaton, Spencer Pitman, Stefano Mazzocchi, Wesley Beary, and the rest of the FLUX Collective

---

We’re a ragtag band of systems thinkers who have been dedicating our early mornings to finding new lenses to help you make sense of the complex world we live in. This newsletter is a collection of patterns we’ve noticed in recent weeks.

“Is it not miraculous, reader, the power of the mind to believe and not believe at once?”

— Ada Palmer, Too Like The Lightning

🧪💡 The prompt test

“Make it better!” you tell the AI. You get something, something surprisingly good, TBH... but it’s not what you had in mind. You need to be more specific. Better how? Better for whom? How does it verify? Most folks who have been using AI awhile know that they have to craft a good prompt to get good results.

Now let’s look in a slightly different direction: many leaders communicate with people in ways that would obviously fail as AI prompts. “Increase velocity.” “Be more customer-focused.” “Drive innovation.” These directives feel meaningful, but they are not actionable. However, unlike LLMs, which fail quickly enough for you to correct (sometimes painfully so), people are really great at guessing. And, right or wrong, those guesses compound as they cascade through organizational layers.

This hints at why AI hasn’t transformed large organizations yet. If all we had to do was get the work done faster, then transformation would be easy. But before we get there, we have to solve the problems of dispersed ownership and vague goals. AI can accelerate coding, writing, and analysis. It can’t resolve disagreement about what should be built. It can’t do meaning-making.

Lafley and Martin’s Playing to Win describes strategy as an integrated cascade of choices: where to play, how to win, what capabilities to build, and what systems to put in place. Each level provides context for the next. “Be more customer-focused” fails because it skips the cascade: which customers, competing against whom, with what tradeoffs, measured how? Effective direction-setting is clear and, just as critically, it’s coherent.

These days, we’re deploying AI at the nodes of the organizational graph, helping individuals do their tasks faster. Transformation at scale requires redesigning the graph itself: the processes, decision rights, and information flows. That’s slow, messy work. For example, it means confronting why all those status meetings and planning processes exist in the first place, instead of just making them more efficient. Egos will be bruised in the process.

The prompt test offers a diagnostic: would you give more detail if you were working with an LLM? If not, what’s missing? The context? The constraints? The actual goal? Over time we will see that AI’s biggest contribution won’t be doing the work faster. It will be exposing how unclear organizational communication has been in the first place.

🛣️🚩 Signposts

Clues that point to where our changing world might lead us.

🚏🇬🇧 Britain will encourage households to use more electricity this summer

In a few months, Great Britain could see “the first summer the grid runs entirely on zero-carbon electricity.” There will, in fact, probably be excessive solar and wind power, and this poses its own problem to grid operators: shutting down wind and solar plants when demand is low is quite expensive. So, the national grid operator will urge Britons to boost their power usage when the grid is over-supplied, such as by doing less time-sensitive tasks like running dishwashers or charging EVs. Some suppliers may offer discounted or even free electricity during these periods.

🚏🪫 An electric car factory will run on its own recycled batteries

Electric car startup Rivian has inked a deal to install over 100 used batteries from Rivian cars in its Illinois manufacturing plant. Besides addressing the problem of EV battery recycling, this will help the plant store energy when grid prices are low and discharge it at peak times, a method known as peak shaving. Indeed, as AI data centers drive up demand for energy storage, recycled car batteries offer a cheap alternative. (While EV batteries do degrade, they degrade less than people initially thought, and even a battery with just 50–75% of its original capacity is still useful enough for applications like these.)

🚏🍁 The number of Canadians visiting the US by car is down 35% since March 2024

Ever since the US began its Canadian annexation campaign in 2025, Canadians have mounted a remarkably long-lived boycott of American travel. The number of Canadians driving to the US (the most common method for them to visit) is down 35% since March 2024; there has also been a 14% year-over-year drop in the number of air travelers from Canada to the US. That has accounted for at least $4.5 billion in missing tourist spending, which has hit border areas hard. One travel consultancy CEO was impressed at the scale of the boycott, saying, “In my 37 years in the travel industry, I have never seen anything like what the Canadians have pulled off.”

🚏🦞 Maine will become the first state to pause data center construction

The boom in data centers, and the resulting energy price rises in their areas, has been drawing backlash across the US, and Maine has become the first state to enact a ban on data center construction. The moratorium, which drew bipartisan support, will last until November 2027. (The governor retains the right to veto the plan before it officially becomes law, though.) Maine doesn’t currently have a major data center project, though some small facilities are under construction or being planned.

📖⏳ Worth your time

Some especially insightful pieces we’ve read, watched, and listened to recently.

• Secrets of Intelligence Services (Ben Recht) — Shares what the author learned from having Claude Code create a minimal coding agent. He observed that, while LLMs are incredibly “mysterious” and “nondeterministic,” agents themselves are very simple and rule-based. In fact, a micro-agent with just three tools (read file, write file, and run terminal command) plus an LLM was able to do most of what a ‘real’ coding agent can — a classic example of emergent behavior. The author concludes that the simplicity of agents in feedback with the complexity of LLMs is what makes coding agents “the most useful and engaging AI product yet.”

• The Monks in the Casino (Derek Thompson) — Argues that young men are increasingly retreating into a toxic asceticism, trading pro-social milestones for porn, posting, and pervasive sports betting. Thompson argues that the true crisis isn’t loneliness, but an alarming comfort with being completely alone, a trend that echoes Ivan Illich’s “two watersheds,” where the virtue of monasticism has mutated into a vice under casino capitalism.

• What the Death of Direct File Tells Us About State Capacity (Don Moynihan) — A public policy professor examines the rise and fall of the IRS’s free Direct File tool, which was piloted in 2024 but killed in 2025. Governments definitely have the state capacity to build software to serve their own ends (such as surveillance tech), and they can turn that to building public goods too, but without broad-based public support, it’s easy for private entities with a financial interest to kill them off. Indeed, the US government shuttered Direct File after a meeting with lobbyists from the tax prep industry. On the bright side, Direct File was a great example of the user-centric, product-forward, in-house model of software development that civic technologists have been promoting as an alternative to external contractors, and it may return in the future.

• The Safest NFL Draft Pick Is the Most Dangerous (Michael MacKelvie / YouTube) — Applies finance’s Sharpe ratio (a measure of how much excess return you get per unit of risk) to football drafting, finding that the common strategies of picking the “best player available” or filling a position of need aren’t actually the most effective. It’s better to apply the lens of replacement cost and take bigger swings at “premium” positions (like quarterback or edge rusher) that are more expensive to get in free agency. The corollary is that taking even a blue-chip player at a cheap-to-replace position (like running back) high in the draft is a pitfall.

🔍📆 Lens of the week

Introducing new ways to see the world and new tools to add to your mental arsenal.

This week’s lens: plausible deniability.

In Disney’s animated Cinderella, Lady Tremaine says Cinderella can go to the ball:

Lady Tremaine: Well, I see no reason why you can’t go... if you get all your work done.

Cinderella: Oh, I will. I promise.

Lady Tremaine: And, if you can find something suitable to wear.

Cinderella: I’m sure I can. Oh, thank you, Stepmother. [Cinderella leaves]

Drizella: Mother! Do you realize what you just said?

Lady Tremaine: Of course. I said, “If.”

Drizella: Oh! “If.”

And when Cinderella seems to have met the conditions? Well, Lady Tremaine never goes back on her word, yet Cinderella somehow ends up with a dress in tatters.

Plausible deniability is a form of ambiguity. It allows people to keep a distance from the role they played in some outcome. Generally seen as a negative, plausible deniability is often present in scandals, both political and social. A politician who expresses their annoyance at an opponent may not say that they want that opponent’s reputation smeared. If their underlings chose to interpret it that way, who can blame the politician? A government that executes an assassination and hides the involvement of the state can turn around and blame others for the death, as could a leader who stochastically asks why nobody will “rid me of this turbulent priest.”

Despite the negative association, it may be more accurate to see plausible deniability as a tool for negotiating weakness rather than as something inherently evil. Our hypothetical politician and government resorted to plausibly deniable actions exactly because they don’t have the power, legally or socially, to execute them directly.

Applied more broadly, we can start to see ways plausible deniability can be positive. One example is flirting. Attempting to establish interest in another person is socially fraught, so good flirting escalates slowly. A look. If that is returned, some eye contact. Then a conversation. At any step along the way, if it goes well, we can keep going. If it doesn’t, we can bail, “Oh, I was just saying ‘hi.’” And like with political plausible deniability, the denial doesn’t always need to be truly plausible to fulfill its role in signalling.

Early conversations around mergers and acquisitions often have this shape. Potential acquirers and acquirees engage in a slowly escalating expression of interest to reveal how serious they are and if there’s a good match. Diplomats use plausible deniability all the time. They suss out who supports their position and who is against it. They deescalate tensions without their government losing face. One could say a significant part of a diplomat’s job is to ensure their goal is achieved without abandoning the official line.

Regardless of the end to which it is used, plausible deniability shows up when being wrong is costly, when we can’t just mandate the outcome, and when we need to preserve relationships. So is it good or bad? We’d say (with plausible deniability of course) that it all depends on the context and the intent.

---

© 2026 The FLUX Collective. All rights reserved. Questions? Contact flux-collective@googlegroups.com.