Sunday, August 2, 2026

Good Sunday, NOLA. Today’s brief is about putting AI to work with a little more judgment: an MIT Sloan study finds that the quality of AI financial guidance depends heavily on the question you bring it; Datasette Apps is making agent-built internal tools easier to debug; and Surfsense is pitching a useful alternative to closed-web research. Also worth a look: practical prompt habits, a few lightweight tools, and the increasingly real question of how AI-made media shows up in public culture.

Use AI With Better Inputs

AI financial advice gets much better when you ask better questions

An MIT Sloan look at AI financial guidance lands on a broadly useful lesson: the tool is often only as helpful as the context, constraints, and follow-up questions you give it. Treat an AI answer as a starting point for comparison and verification—not a substitute for professional advice. The Hacker News discussion is a good companion read for the practical caveats.
MIT Sloan / Hacker News

Ten ways to get better AI answers

A handy weekend refresher on the highest-leverage habit in AI work: give the model a role, useful context, a concrete output format, and room to ask a clarifying question. The best prompt upgrades are usually less about clever wording than making your actual goal legible.
There’s An AI For That

Surfsense wants your research to reach beyond the usual AI answer box

Surfsense is framed as an open-web search tool—a useful direction for anyone who wants AI help finding sources instead of receiving a polished but opaque summary. Worth trying on a topic where you already know enough to judge the results.
There’s An AI For That

Small Builder Upgrades

Datasette Apps adds a debugging tool for agent-built apps

The new app_debug() tool lets an agent open and inspect an app while it is being built or edited. That is a modest feature with a very practical payoff: less guessing when an AI-generated internal tool looks right in code but not in the browser.
Simon Willison

A reminder for teams putting ChatGPT-style helpers in Slack

Simon Willison highlights an important social-design rule: people may welcome an AI helper, but not an unexpected bot message from a coworker’s account. If you are building a workplace agent, make its identity, permissions, and handoffs obvious from the start.
Simon Willison

Serko.ai promises a one-prompt shortcut for routine work

One-prompt automation is an ambitious promise, but these are exactly the tools worth testing against a real recurring task. Pick something small, check the output carefully, and see whether it saves a meaningful step—not just a few clicks.
There’s An AI For That

Gemini Spark is being pitched as a bigger Chrome companion

Browser-native AI is where a lot of everyday workflow experimentation is happening. The interesting test is simple: does it remove the annoying copy-paste loop between your tabs and your chat tool?
There’s An AI For That

Snapchat is dialing back low-quality AI content

The product signal here is bigger than one platform: audiences are getting sharper at noticing low-effort generated material. For builders, provenance and taste are becoming product features—not finishing touches.
There’s An AI For That

Culture & Business Signals

When an AI-content question reaches the Billboard chart

The Verge examines a charting hit surrounded by questions about how it was made. For creative teams, this is the live version of a coming product decision: be clear about where AI was used, because audiences will increasingly ask before they commit their attention.
The Verge

Worth a listen: the AI investment story is getting more concrete

AI Daily Brief follows Microsoft’s view that AI spending remains durable. It is a business-side listen, but useful for builders trying to understand why the major platforms keep turning AI features into default parts of everyday software.
AI Daily Brief

Today’s Sources