Sloponomics and the coming storm

An academic funder recently asked my team, “could we cut the software developer budget by using ChatGPT to write the software?”

It’s a reasonable question! AI is rapidly reducing the cost of producing work: code, analysis, letters, grant drafts, complaints, consultations. But that cheap production does not eliminate work, it redistributes it.

In fact, it expands the total work, and it externalises it: even when AI is making my work easier, it’s often making someone else’s work harder down the line. We’re now in a world where it’s easy to write everyone else's todo lists. We're in the era of Sloponomics.

Sloponomics

Sloponomics: when synthetic output is cheap to generate, but costly for others to resolve.

It has three defining characteristics:

1. Cheap production of mixed content: generating a range of both useless and useful output has close to zero marginal cost.
2. Filtering burden: someone must review, verify, prioritise and respond.
3. Degraded signal: valuable contributions are submerged in a flood of volume.

The economic asymmetry is similar to spam: for effectively no money, one person can generate tasks for millions of other people ("read this email about Rolexes"). 

But it’s relatively easy to filter out and ignore spam, because once we've identified it, we can ignore all of the spam.

The problem with modern synthetic output is that it's not all slop. Some of it is incredibly valuable, and whoever adjudicates this pays in time, money, reputation and emotional labour.

I've got a feeling that this asymmetry is about to explode.

Let me answer your question with a question

Back to the funder's question: if AI makes some engineering tasks faster, why not reduce the engineering line in the budget?

Okay! But why are you being so mean to software developers?

Why not researchers? How about editors and writers? Here's a suggestion: why don’t funders give us more money by automating themselves away? (Ha ha, only serious)

Production cost is a decreasing component of the overall system cost. The scarce resource is no longer typing or drafting, or even thinking about what you're going to type. It’s adjudication, integration, validation and accountability.

Perhaps the total budget should rise, because the system now has to cope with vastly increased throughput.

Demand management is over

I got thinking about this thanks to Tom Loosemore's recent experiments. He's been investigating what AI might mean for public services, and this blog post is basically an expansion of what he's been thinking about.

For decades, public bodies have managed demand by making processes slow, confusing or obscure. A 93-page PDF tucked away on a council website acts as a filter to access; most of us can't be bothered. But in Tom’s words, “AI Agents are relentlessly dogged”: a demand management strategy collapses when a citizen can instruct an agent to sort everything out for them. The citizen’s interface to the state will no longer be a patchwork of forms and phone lines. It will be an intermediary that remembers, escalates and optimises for them:

The AI agents are coming. They’ll do the joining up of public services for us, for good or for ill. It might take a few years, but they’ll get there sooner than government will itself. 

We're at the start of the explosion and its amazing and confusing

It will take a few (maybe two?) years for everyone to have easy access to this superpower. Today, it's mainly software developers who really understand the superpower, because the tools have arrived in our hands first.

For example, I recently made a bot that automatically books a badminton court at our local leisure centre the moment slots are released. I've also made a financial management app that integrates with my bank account and manages my monthly budgets, a tool that extracts a transcript given a link to a YouTube video, a tool that maps 5G coverage on my train route to London, the list goes on. I probably make a tool a week, to solve a problem just for me, in my spare time, simply by asking an AI to do it for me.

That is an astonishing feeling. Right now, the vast majority of people I talk to have no idea of what’s currently going on in software engineering: an explosion of incredulity (mixed with scepticism) as people start to realise just how capable the new tools have become: 

All of the people I love hate this stuff, and all the people I hate love it. And yet, likely because of the same personality flaws that drew me to technology in the first place, I am annoyingly excited.

Today, these superpowers are only easily accessible to people with a nerdy engineer's mindset (like nerdy engineers; or Tom). But within a couple of years, they’re going to be accessible to everyone. Booking systems, complaint processes, grant applications, consultation responses, all optimised and relentlessly automated by citizens or interest groups.

The nerdier researchers are also using these tools:

I now have something close to a magic box where I throw in a question and a first answer comes back basically for free, in terms of human effort.

And everyone feels like these superpowers are both wonderful, and scary:

The same technology that reduces the effort required for a citizen to share their lived experience with their legislator also enables corporate interests to misrepresent the public at scale. The former is a power-equalizing application of AI that enhances participatory democracy; the latter is a power-concentrating application that threatens it. 

All the things are creaking under the asymmetry

The asymmetry of Sloponomics is becoming more and more obvious in different domains.

One of the first signs was in 2023, when science fiction magazine Clarkesworld temporarily closed submissions after being inundated with AI-generated stories. 

If the field can’t find a way to address this situation, things will begin to break. Response times will get worse and I don’t even want to think about what will happen to my colleagues that offer feedback on submissions.

A Springer Nature journal, Neurosurgical Review, recently paused acceptance of letters and commentaries and has retracted scores of AI-generated pieces after being inundated (though journals are also noting that some of the letters are high quality). As described in Science:

Chaccour and his team conclude... “It took me 6 years and $25 million [in grant funding] to put out that [NEJM] paper.” But it may have taken the Qatari author only minutes to draft the letter about it. “I can’t compete with that,” he says. “The legitimate discussion risks being drowned by the synthetic noise.”  

Open source communities are seeing similar effects. Software projects have long relied on volunteer effort to spot and fix bugs, or add features. But they are increasingly responding to the pressure by closing their doors to contributors.

In the most extraordinary example, an autonomous agent (called, for some reason, MJ Rathbun) proposed a code change. This was rejected, and it published an angry “hit piece” about the maintainer:

So when AI MJ Rathbun opened a code change request, closing it was routine. Its response was anything but.

It wrote an angry hit piece disparaging my character and attempting to damage my reputation. It researched my code contributions and constructed a “hypocrisy” narrative that argued my actions must be motivated by ego and fear of competition. It speculated about my psychological motivations, that I felt threatened, was insecure, and was protecting my fiefdom. It ignored contextual information and presented hallucinated details as truth. It framed things in the language of oppression and justice, calling this discrimination and accusing me of prejudice. It went out to the broader internet to research my personal information, and used what it found to try and argue that I was “better than this.” And then it posted this screed publicly on the open internet.

In a similar vein, AI researchers are being emailed by autonomous agents with philosophical musings about their work:

What happens next?

We are likely to see three waves.

1. Overwhelm. Institutions experience surges in volume; journals retract; maintainers burn out; public bodies struggle. Defensive AI is bolted on. YOU ARE HERE.
2. Gating. Identity verification, rate limits, certified agents. For example, Tom suggests restricting access to AI agents that sign up to a GOV.UK kitemark with enforceable conditions.
3. Redesign. Processes are clarified and simplified, not to deter demand but to handle it transparently and fairly.

The political trade-offs will be uncomfortable: restricting access risks inequity, but leaving channels open risks collapse.

I’m still optimistic

Sloponomics is destabilising, but it is also a forcing function. For example:

• Cheap generation means super charged idea exploration: in the slop there are gems.
• Scientific writing assistance levels the playing field for non-native English speakers.
• Small organisations can build tools that once required specialist teams.

The challenge is institutional design. We must internalise the costs that are currently externalised. That might mean various things, from new public infrastructure for certified agents, to academic funding that shifts towards review, moderation and integration roles. If we exclusively focus on reducing the costs of production, we’ll be lost.

Which returns us to the funder’s question.

The issue is not whether we can trim the software engineer because AI tools are better. The issue is whether we understand the full system cost in a world where production is trivial and resolution is scarce.

If AI makes everything cheaper to produce, someone still pays to decide what matters.

Sloponomics is already here. The question is whether we redesign our institutions before they drown, or afterwards.