Sponsored by

MediaMorph Edition 110 - by Mark Riley

Where did all the ‘free’ time go?

Was this newsletter forwarded to you? Sign up here

Welcome to our new readers from New Statesman, Storyful and Reach

The written-by-a-human bit

Adam Smith, Karl Marx and John Maynard Keynes all warned about the excess of needless work and offered us the promise of shorter working weeks. Adam Smith told us in The Wealth of Nations that long, monotonous labour can render workers “as stupid and ignorant as it is possible for a human creature to become,” because they have no time or energy for education, reflection, or civic life. Karl Marx, in Capital, extols that only when the working class wins a shorter working day do workers gain time for education, political activity, and the “development of human energy which is an end in itself.” Keynes famously promised us in his 1930 essay “Economic Possibilities for our Grandchildren” that, within about a hundred years, productivity growth in advanced economies would be so high that people would need to work only about 15 hours a week to satisfy their material needs.

Instead, in the age of AI, despite the promised “hours saved,” we seem to be working longer hours and more intensely than ever.

Welcome to the age of “botsitting, botshitting, and the hidden human labor of AI at work”. (Thanks to NLW at The AI Daily Brief for the insight).

AI platform provider Glean and The Work AI Institute have released a report that suggests that 87% of digital workers use AI, 75% say it makes them more productive, and the average worker claims to save 11 hours a week through automation. And yet only 13% say their organisation is performing significantly better.

That gap is the story. The report gives it a useful name: “botsitting” — the hidden labour involved in making AI actually useful. It is the time workers spend feeding tools with missing context, checking outputs, correcting errors, re-running prompts, and comparing answers across different systems before those answers reach the real world.

Apparently, on average, workers now spend 6.4 hours a week botsitting. That is 37% of their AI time — more than they spend actively using AI itself. Some of this is productive: reviewing high-stakes outputs, adding domain expertise, refining prompts and learning how to work better with the machine. But a lot of it is pure waste: reloading the same information into different tools, chasing better answers, debugging hallucinations and compensating for systems that do not understand the work well enough.

I have found myself working late into the night fixing a vibe-coded website that always seems to have one last bug to sort - apparently it is the same dopamine hit that slot machines provide.

Worse still, “Botshitting” is when people become so used to AI producing plausible-looking work that they stop properly checking it and start passing it on as if it were reliable.

It is not just “AI hallucination”. The hallucination is the machine’s error. Botshitting is the human failure to verify it.

Laziness, indifference and verbosity are the new problems to watch out for. Know anyone?

Glean’s solution is to stop treating AI adoption as a sequence of bolting on new tools but as a new operating system. The most successful teams keep humans accountable, use AI as a teammate rather than an oracle, and spread adoption through peers rather than mandates. They also suggest tracking quality, output, productivity, and time saved, not just tool usage (and presumably buying Glean).

There are other solutions - go through the pain of a one-time context dump for Claude or Codex and build reusable AI-assisted workflows (we can help).

We are way past AI being a technical challenge. It is now a cultural, human, and behavioural challenge. Using AI well is both a mindset and a skill set.

Speaking of getting AI wrong, my weekend longread was "Europe 2031 - What getting AI wrong means for us" (discovered via the excellent Times Tech Podcast).

It is an explosive narrative wake-up call to the EU politicians and bureaucrats to stop talking and start building AI data centres. The authors, including Judith Dada, take us on a five-year journey to 2031 through the eyes of fictional characters: Caroline Dubois, a French policy worker in Brussels, and Christian Vogt, the German founder of a fast-growing AI company who has just relocated to Silicon Valley.

It relates how, without compute, the EU will be ripped apart by competing pressures from the US and China and slide into irrelevance. (One sentence made me laugh: “and some have started talking about moving to the UK, which has - to the surprise of many - manoeuvred the AI transition much better than most EU countries have”.)

Apart from being a cracking read, there is a media story here. The paper caused an enormous splash and is being widely read by politicians and policymakers. The authors are all AI researchers, think-tankers, and investors, who brilliantly executed a compelling narrative for outsized impact. Journalists should take note.

Mark Riley, CEO Mathison AI

Take our 10-minute AI Readiness Audit now for free benchmarking against your peers

Book an AI masterclass or workshop now - [email protected]

AI and Journalism

This week’s best articles, as chosen by Hana and our editors

Conversation on AI and Journalism • Christo Buschek

Gijn -

In a masterclass video, Christo Buschek, a programmer and investigative journalist from Der Spiegel, explores how programming skills can enhance journalism by facilitating data analysis and promoting transparency. He offers practical advice for aspiring journalists on integrating technology into their storytelling while addressing ethical considerations related to data privacy and the responsibilities towards sources.

The best voice models, now across all channels

Most CX platforms do not own the voice. They orchestrate a workflow, then call a third party for speech and transcription. Every hop adds latency, cost, and another vendor to manage.

ElevenAgents is the opposite. They make the voice models the market builds on, and ElevenAgents puts full orchestration on top. Voice, transcription, text-based chat, and reasoning run in one vertically integrated pipeline, so responses come back in <400 milliseconds and sound human, not synthetic.

Plus, you keep full control. Plug in any LLM, integrate tools, webhooks, and MCP servers, and ground responses in your knowledge base. Get an agent live in minutes, then A/B test with Experiments, enforce Guardrails, and version every change.

The payoff: more human conversations, lower latency, and far less time stitching infrastructure together. You build on the models you already trust. Pricing is transparent and flat at $0.08 per minute.

Start building free

AI and Academic Publishing

This week’s best articles, as chosen by Hana and our editors

AI promised to democratise academic publishing – the evidence says otherwise - LSE Impact

Lse -

While AI tools can assist non-native English speakers in refining their academic writing, they often overlook deeper issues like cultural context and biases in the peer review process, which continue to disadvantage these scholars. Ultimately, reliance on AI may compromise the authenticity of individual voices in academia, highlighting that improved language skills alone do not create an equitable publishing landscape.

MediaMorph Edition 110 - by Mark Riley

MediaMorph Edition 110 - by Mark Riley

Where did all the ‘free’ time go?

The written-by-a-human bit

The best voice models, now across all channels

Keep Reading

MediaMorph

MediaMorph