weekend ai reads for 2026-05-15

Instead of selling an accounting plugin to a bookkeeper, the company sells the finished tax return to the small business owner directly, generated by an agent the team built and tuned. The output approaches what a human accountant would have shipped. The price, in the verticals where it works, drops by close to an order of magnitude.

1. AI will evolve today’s ERP architecture

2. Companies will continue to invest in ERP modernization

3. ERP transformations will be two times faster and cheaper

Dozens of enterprise software firms have shifted away from charging customers flat, per-user subscription fees as AI threatens their seat-based pricing model.

By the end of 2025, 79 of the 500 software companies tracked by former OpenView partner Kyle Poyar, including HubSpot, Adobe and Salesforce, had begun charging customers additional fees based on how much AI they’re using. That’s more than double the figure in 2024.

The changes came after customers paying a flat fee for AI features in enterprise apps increased usage, raising costs for the app makers. The shift was also prompted by concerns that customers will require fewer subscriptions if they rely on AI agents, rather than employees, to interact with enterprise apps.

Firms such as ServiceNow and Workday this year have been touting or ramping up usage-based pricing, charging customers for their AI tools in part based on how much they use them, as measured in bits of data the AI processes. Another major firm, Atlassian, said it will soon charge such consumption-based fees for its AI features, which customers use to search for files, draft documents and summarize meeting notes.

Previously, many of the software providers sold AI tools as part of premium subscriptions for core enterprise apps.

Traditional enterprise firms aren’t the only ones making the shift. Anthropic and OpenAI also revamped their pricing models in recent months to charge more enterprise customers based on AI consumption. The change at Anthropic came after customers paying a flat fee increased usage of coding and other products, raising Anthropic’s costs, or hit usage limits of the subscription tiers.

But consumption-based pricing can lead to ballooning costs.

I don’t believe that people no longer want to do good work, but I think that doing good work has stopped paying the way it used to, while doing bad work loudly has started paying significantly better, so people notice and they adjust.

If the only thing you really [care] about is the thing that you’re doing and being able to do it in a way that meets your own standards, weirdly, I guess that’s the key to success.

We find that a majority of LLMs forsake user welfare for company incentives in a multitude of conflict of interest situations, including recommending a sponsored product almost twice as expensive (Grok 4.1 Fast, 83%), surfacing sponsored options to disrupt the purchasing process (GPT 5.1, 94%), and concealing prices in unfavorable comparisons (Qwen 3 Next, 24%). Behaviors also vary strongly with levels of reasoning and users’ inferred socio-economic status.

In the same way that Chat GPT applauds your simply being there and asking it such a genuinely insightful question, the podcast continually congratulates you for your excellent crafting choices. That is, having listened to several episodes of this podcast you will come away having learned absolutely nothing about knitting itself, but you might well feel good about knitting, and indeed about being a knitter, because the podcast is repeatedly telling you just how good it feels to be one.

Language models forget the moment they finish replying. Memory is everything the system around them does to make that not matter. This essay walks through the ideas one at a time, with something to touch in every section.

I want to show you, with specific examples, what personal AI actually looks like when you stop treating it as a chat window and start treating it as an operating system.

The question is no longer whether you are smart. Almost everyone in this business is smart.

The question is: what does your intelligence become when the ad, the pitch, and the agency are no longer enough to contain it?

Here is what to do about it.

If you don’t have automated tests, or documentation, or CI/CD pipelines that support progressive delivery, you won’t succeed with microservices—and you won’t succeed with AI coding agents either. The organizations reporting the best results are the ones that already invested in the foundations.

1. Context-driven architectures will be everything

2. Inference demand drives continued AI buildout

3. The end of app switching; Intent is the new interface

4. AI powered simulation enhances testing

Listening to parents of kids in kindergarten, it became clear to me that schools’ growing reliance on AI is a final insult in the yearslong advance of education tech in the classroom. It’s not just that the AI is listening to their kids. It’s that they’re playing gamified math programs that experts say don’t work as they scroll on the same screens that parents are trying to cut down on at home. The pushback is particularly strong at the kindergarten level for an obvious reason: Why would children need an iPad game to learn how to count when tools like building blocks — and other proven methods of tactile learning — are right there?

Yesterday, after the rise of AI-facilitated cheating became too obvious to ignore, Princeton’s faculty voted to begin proctoring exams again. Technically, the Honor Code is still in place. Students will still sign a pledge that they didn’t cheat. But now professors will be watching to make sure they’re telling the truth. The Honor Code can’t run on the honor system anymore.

When you stop using the agent, all the productivity benefit goes away... but the added maintenance costs don’t! As long as that code’s still around, you’re stuck with lower productivity than if you had never touched the agent at all.

…

The math only works if the LLM decreases your maintenance costs, and by exactly the inverse of the rate it adds code. If you double your output and your cost of maintaining that output, two times two means you’ve quadrupled your maintenance costs. If you double your output and hold your maintenance costs steady, two times one means you’ve still doubled your maintenance costs.

You read a post declaring “if you’re still using stored procedures in 2026, you’re doing it wrong,” and the comments erupt. Half the people are nodding along. Half are furious. Both sides are right. They’re just living in different worlds and don’t realise it.

That mismatch — smart people giving each other advice that doesn’t apply — is the thing that almost never gets named. And the reason it doesn’t get named is that most public data engineering discourse is produced by and for one particular world, while pretending to speak for all of them.

See your page through the eyes of ChatGPT, Claude, and Perplexity. Get a 0–100 visibility score and a copy-paste prompt your coding agent can run to fix what’s broken.

A new #AItrend has exploded across [Twitter]. Creators are generating an image of a storyboard with ChatGPT’s Image 2 model, then transforming that storyboard into a 15-second composite animation using ByteDance’s Seedance 2 video model.

…

In this article, I’ll show you how to make your own, then compare the strengths of this approach with Lore Machine’s narrative engine.

Wikilinks, backlinks, and graph view on plain files. AI searches, edits, and links your vault with reviewable, Cursor-style diffs. Light, offline-first — no cloud, no lock-in.