The Art of Agency Laundering

🍕 I’m JA Westenberg. I write essays about technology, philosophy, productivity, and what it means to be human. If you like ideas that take their time, you’re in the right place.

While you’re here, a few things I’ve built: Kerouac, a writing app | Post Through It, a solo microblogging tool | Distributism.ai, my MCP platform | Laterware, all the AI news you need | Studio Self, a thought leadership agency for founders x VCs.

You can buy my book Permissionless here.

Everyone has some version of the same trick.

You don’t want to take responsibility for a Great and Terrible Thing - your choices, your decisions, taking a risk, playing it safe, etc.- and so you attempt to remove yourself from the causal chain. Best-case scenario, you flip a coin. Bad-case, you consult a psychic medium.

Worst-case, of course, you hand it all off to a Board...

Whatever you choose to do, the intention is the same: you are inserting an additional link in the causal chain, between you and the result; and you’re assigning the moral weight, blame, and absorption of whatever comes next to that link. It’s no longer your problem. It’s no longer your fault. You’re blameless in all of this, because there is something or someone else, be it chance, the spirits, or various semi-retired executives, who are there to take the fall.

Humans do this. Software companies do this. Governments do this.

I do this.

You do this.

The newest iteration of it is agentic agency laundering.

How’s that for a sentence?

It’s the cleanest form of the contract. AI can’t recuse itself or accuse you. If you delegate your decision-making to Claude, it won’t argue, fight back, or write a tell-all memoir. And unlike the spirit world, it’s at least somewhat more concrete in its responsiveness. It also has a bit more weight and aura. Nobody likes being laid off, but if you can claim a model determined it was the Right Call Given the Circumstances, it sounds like a degree of high-tech wizardry went into it, and the headlines might be a little more forgiving.

There’s less expectation of general humanity, for that matter.

Of course, a model denied that health insurance claim; models have no empathy. What can you expect?

All of which works, up to the limits of your fellow man’s gullibility: the extent to which they’ll happily forget that no matter what the model did, presumably at some point, you requisitioned it, and you paid for it, and you gave it the workload and a degree of authority and let it loose.

The moment they do remember that, though, one can only assume that all hell will break loose.

That’s the problem with agency laundering.

You can never really obfuscate where the chain begins.

We have a long, if not entirely distinguished, tradition of adding an intermediary, assigning blame to it, and then running away from the result. Pontius Pilate had a basin of water, Croesus consulted Delphi, and Macbeth listened to three women in a fog who told him what he already wanted to hear.

The witches, after all, didn’t make him murder anyone.

They were simply an inserted link in the causal chain that gave him a plausible level of agency laundering…

A coin toss is a terrible alibi, a psychic is a walking extortion risk, and a committee just spits out bureaucratic noise. But an algorithm hands you the perfect shield: a slick, formatted memo loaded with caveats. It takes about 40 seconds, and you can automate it so it goes from being birthed in one Slack channel to being anointed in another.

Meaning decision-makers can abdicate every Captain’s Call to the most polished prose machine on the planet...

Alan Rubel, Clinton Castro, and Adam Pham called the formal version “agency laundering” in 2019. An agent who already bears moral responsibility inserts a process into the causal chain, credits that process with neutrality, expertise, or method, then requires that the process answer for the result.

When agents insert technological systems into their decision-making processes, they can obscure moral responsibility for the results. This can give rise to a distinct moral wrong, which we call “agency laundering.” At root, agency laundering involves obfuscating one’s moral responsibility by enlisting a technology or process to take some action and letting it forestall others from demanding an account for bad outcomes that result.

A committee member can leak, and a consultant can go on a podcast; but a model won’t reveal that you framed the task to guarantee one output or ran the query seven times until the answer matched your preference.

Because it has no stake in the story, you can park a decision there and walk away scott-free.

We’re still early enough that “the model recommended” still sounds at least somewhat methodical; a person who’d roll their eyes at a horoscope might still accept a temperature-sampled paragraph from an acceptable name-brand AI model.

Which won’t last - but more skepticism won’t end the practice on its own...

Credulity helps, but the person seeking cover wants the intermediary even when no one believes in the ritual. Pilate wanted the basin because he needed a public gesture that placed the blood elsewhere, not because water could absolve him.

The architecture of modern accountability is deliberately flawed. Madeleine Clare Elish defined this in 2019 as a “moral crumple zone” - a framework in which low-level human operators absorb the liability for structural failures.

What is unique about the concept of a moral crumple zone is that it highlights how structural features of a system and the media’s portrayal of accidents may inadvertently take advantage of human operators (and their tendency to become “liability sponges”) to fill the gaps in accountability that may arise in the context of new and complex systems.

These operators are designated as safety drivers but are systematically denied the operational authority to alter course. The institution prioritizes the system's and its executives' survival by scapegoating a disempowered intermediary.

The Nuremberg Charter historically invalidated this dynamic, establishing under Article 8 that acting as a mere instrument or following directives does not grant immunity, regardless of the duress involved.

I’m not implying that a product launch belongs beside a war crime on the scale of human ills (despite how I feel after scrolling Twitter for more than a handful of minutes) but the narrower point is that delegation doesn’t erase the person who began the chain.

To my mind, the correction is simple: keep your place in the chain visible.

If you use a model to draft your rationale, you have to disclose it. If you use an agent to help you write, you have to share the credit.

As the tech gets smarter, the excuses get lazier. Corporate suits are rebranding automated scripts as “autonomous workers” just to launder their liability. They want you to believe the software went rogue on its own, with no headcount to blame; don’t buy it.

Every disaster points straight back to the executive who signed the check, set the target, and rubber-stamped the output.

When the fallout hits, the human who pulled the trigger will still have to own the wreckage.

The code is not your meat shield.