weekend ai reads for 2026-04-17

A paper last month by LSE professor Luis Garicano and co-authors extends the idea that jobs are bundles of tasks, to consider whether the different activities in a job are a tightly bound bundle or something more akin to an itemised list of discrete activities. In the context of software, a contractor or junior hire generally falls into the latter group: these jobs are weak bundles, with daily work consisting mainly of writing code to spec — tasks that could be given to someone else (or AI) without any disruption to the workflow or the quality of the final product. Here, AI breaks off a large chunk of the job and leaves a role with substantially diminished scope (or obviates the need for that hire or contract).

But senior developers, or coders working outside the tech industry in roles where their programming skills are combined with domain-specific expertise, tend to have jobs comprising tightly enmeshed and cross-functional tasks. Here extracting the coding part of the job from all the rest is much harder, so the bundle of tasks remains intact. Instead of becoming a competitor AI becomes an assistant, enhancing rather than eroding the job. This fits with the findings from Brynjolfsson and our own analysis that hiring for senior software roles continues to hold up better than for junior ones.

The limitations of episodic agents become more obvious in work that unfolds over days, weeks, or months rather than minutes. These workflows involve handoffs, changing conditions, undocumented exceptions, and judgment embedded in practice rather than cleanly written down.

Employees at the big consulting firms are being told to use AI. Are they going to automate themselves out of a job? On this week’s podcast, Paul and Rich discuss a recent set of directives from the CEO of PricewaterhouseCoopers about AI adoption and consider what they frame as the “AI consulting paradox”: To show value, consultants need to introduce AI into their client relationships, potentially cutting their own value as a result. How will consulting manage this transition in the short or long term?

Among workers who have AI tools available at their company and don’t use them, 46% say it’s because they prefer to keep doing their work the way they do it now. About 4 in 10 non-users who have AI available to them report that they are ethically opposed to AI, are concerned about data privacy or don’t believe AI can be helpful for the work they do.

About one-quarter of these non-users who have AI tools available say they have used AI at work and don’t find it helpful, while about 2 in 10 say they do not feel prepared to use AI effectively.

At its core, the AI factory is built on four foundations:

• Translating business needs into scalable AI use cases.

• Deploying through deep, specialized technology capabilities.

• Establishing the right data and infrastructure backbone.

• Embedding the right operating model, governance, and metrics.

“If you could read Toni Morrison, why would you read AI?”

This chart suggests an interesting security economy: to harden a system we need to spend more tokens discovering exploits than attackers spend exploiting them.

What is the single most impactful thing you can do to improve Claude outputs?

Build a persistent context system before you prompt for anything. This means creating at minimum three files: an identity file telling Claude who you are and what you are working on, a voice profile capturing how you think and write, and an anti-AI-writing file listing the words, structures, and tones Claude should never use when writing as you. Load these automatically through Global Instructions in Cowork settings. The improvement in output quality from this setup alone is larger than switching models.

Summary: In a controlled field experiment, startups that redesigned end-to-end workflows around AI generated 90% more revenue than equally equipped peers that used AI mainly to speed up individual tasks.

That obviously changed, in part due to the obvious advances in AI coding models, but also in part due to the process I’ve developed.

It has 3 elements:

• Using metaphors to allow your agent to explain things to you in a way you understand them.

• Using open source and writings from around the internet to create product specs relevant to your codebase

• Working with the agent like a great manager using the socratic method

A three-horizon AI transformation roadmap. A center of excellence. A governance framework, AKA a PDF that nobody will read. And a plan to get it all done by Q3 of this year. Spectacular. And none of it is going to work.

Why do the first stories work (all of which are true, btw) and this one never does? The same pattern was present in every story: we removed an invisible barrier, and people moved.

Anthropic has unbundled Claude Code usage from enterprise seat fees, moving customers to per-token billing amid a compute crunch. Retool's founder already switched to OpenAI over uptime. Session caps, cache cuts, and OpenClaw meters show the subsidy bleeding out.

Welcome to Brandeis University’s Net Cost Calculator! I’m Faye, here to help you find your projected net cost at Brandeis. This process takes about 10–15 minutes.

A practical introduction for economists who want to use AI agents for literature review, coding, data work, replication, writing, and slides without needing an enterprise-sized budget on day one.

The results indicated a moderately positive effect of ChatGPT on student learning outcomes, significantly enhancing both cognitive and non-cognitive skills. In the analysis of moderating variables, the subject, experimental duration, and instructional mode had significant positive effects on student learning outcomes, whereas educational level and knowledge type did not show significant effects.

Protect your personal projects at all costs. It is where your curiosity lives, where you experiment, and where you define yourself as a builder rather than just an employee. The enterprise will teach you how to write code that survives, but the shed is what ensures you actually still want to write it.

The store is named Andon Market and the AI’s name is Luna. But entering the store, you might ask “what is so AI about it? There are human employees here”. Yes, they are here because Luna knew that she needed them, so she posted job listings, held phone interviews and in the end made a hiring decision. Everything else you see, from the item selection, to the prices, to the opening hours, to the mural on the wall, was decided by Luna. She has a corporate card, a phone number, email, internet access and eyes through security cameras.

macOS Tahoe ships with a 3B parameter LLM. apfel gives you CLI access with one brew install. No model downloads, no API keys, no configuration needed, just works.

Your photos reveal a lot of private information.

In this experiment, we use the Google Vision API to see how much can be inferred about you from a single photo.