Another Crazy Day in AI

Another Crazy Day in AI: Why Some AI Use Helps and Some Hurts

3 days ago
3 min read

Hello, AI Enthusiasts.

Here's another crazy day in AI:

The cognitive crunch nobody's talking about
A new guide for building fair AI in healthcare
You can now plan trips by asking Google Maps
Some AI tools to try out

🎧 Listen to a quick breakdown of today’s stories.

Why Some AI Use Helps and Some HurtsAnother Crazy Day In AI: The Podcast

3:17

TODAY'S FEATURED ITEM: The Mental Price of Peak Productivity

A robotic scientist in a classic white coat with 'AI Scientist' on its back stands beside a human scientist with 'Human Scientist' on their coat, looking towards the AI Scientist.

Image Credit: Wowza (created with Ideogram)

What if the productivity gains come with a cost we're not tracking?

Researchers from Boston Consulting Group and the University of California, Riverside published a study on Harvard Business Review exploring something many workers have probably felt without having words for it. After surveying 1,488 full-time U.S. workers, they found that certain ways of using AI are pushing people to their mental limits — not in the vague, overworked sense, but in a specific, measurable way. They've named it "AI brain fry," and it describes the cognitive exhaustion that builds up when someone spends their day monitoring, managing, and constantly engaging with AI tools at high intensity. The study also looked at where AI use seems to ease the burden, and the difference between the two is more specific than you might expect.

A few observations from the study:

Closely monitoring AI outputs demands more from the brain — workers reported higher mental effort and information overload the more oversight their tools required.
Productivity climbed with two to three tools used simultaneously, but adding more often brought it back down.
About 14% of participants described "brain fry" — difficulty concentrating, slower decision-making, and a persistent sense of mental clutter.
The fatigue showed up most in marketing, HR, operations, engineering, and finance, and least in legal and leadership roles.
Workers who reported brain fry also reported more decision fatigue, more frequent errors, and stronger intent to leave.
Using AI to handle repetitive tasks was linked to lower burnout and higher engagement.
Manager support, structured team practices, and clear organizational communication were all associated with lower mental fatigue.

Mental fatigue from technology isn't a new conversation, but this study puts some harder numbers around something that's been easy to dismiss as people just needing to adapt. The symptoms workers described — the mental fog, the compulsive double-checking, the feeling of managing tools rather than actually doing the work — showed up consistently enough across industries and roles to suggest this isn't just individual sensitivity.

The results also varied enough across roles, team structures, and organizational cultures to make a blanket conclusion difficult. Some workers felt the relief of offloaded tasks. Others felt buried under the oversight those same tools required. Both outcomes came from using the same technology — just differently.

Read the full study here.

OTHER INTERESTING AI HIGHLIGHTS:

Researchers Propose Framework for Fair Health AI

/Patricia Brandt, University of Utah Health Sciences, on Medical Xpress

Researchers at Huntsman Mental Health Institute have introduced SAFE AI, a framework designed to guide the ethical development and deployment of AI tools in health care. Published in the Journal of Medical Internet Research, the approach integrates fairness checks, bias monitoring, and transparency into standard AI development workflows. The framework aims to help organizations—especially smaller health tech companies—identify risks early and ensure AI systems support equitable patient care. Its focus is particularly relevant for mental health settings, where biased algorithms could affect treatment decisions for vulnerable populations.

Another Crazy Day in AI: Where AI Agents Work and Where They Don't Yet

5 days ago
3 min read

Hello, AI Enthusiasts.

Here's another crazy day in AI:

What AI Agents are actually being used for in 2026
Google Cloud expands security with Wiz
Slack adds Claude to improve search and recaps
Some AI tools to try out

🎧 Listen to a quick breakdown of today’s stories.

Another Crazy Day in AI: Where AI Agents Work and Where They Don't YetAnother Crazy Day In AI: The Podcast

4:02

TODAY'S FEATURED ITEM: What AI Agents Can Actually Do in 2026

Image Credit: Wowza (created with Ideogram)

Are your organization's people and data actually ready for what AI can already do?

In this episode of the Sidecar Sync Podcast, host Mallory Mejias breaks down the findings from the 2026 State of AI Agents report by Anthropic and translates what it means for associations and organizations today. Drawing from insights gathered from more than 500 technical leaders, the discussion explores how organizations are moving beyond simple chatbots toward autonomous agents that can handle complex, multi-step tasks—from research and reporting to customer support and software development. The big takeaway: the real value isn’t just saving money. It’s gaining speed, improving quality, and unlocking the knowledge already inside an organization.

Here is what the report found

AI agents complete multi-step tasks independently by working within an organization's existing tools, processes, and data
90% of surveyed organizations use AI agents for coding, averaging around 59% in time savings across development phases
Research and reporting is the top planned use case at 56%, valued for its usefulness and lower risk
System integration (46%) and data quality (42%) are the biggest barriers — not the technology itself
Employee resistance is a recurring theme, especially in smaller organizations, and requires structured attention
Case studies from Thomson Reuters, L'Oreal, and the Norwegian Sovereign Wealth Fund show agents making institutional knowledge far easier to access and use
Most practitioners recommend the same starting point: understand your data first before committing to any use case

The report makes a fair point that most barriers to adoption are not technical ones. Data infrastructure, integration challenges, and internal resistance are the kinds of problems that take time and deliberate effort to work through, and that is true regardless of how capable the technology gets.

What organizations choose to do with that information will vary, but the data is clear enough that it is worth taking seriously. The gap between organizations that are prepared and those that are not is likely to become more visible over the next few years.

Watch and listen here.

OTHER INTERESTING AI HIGHLIGHTS:

Google Cloud Expands Security With Wiz

/Thomas Kurian, Google Cloud CEO, on Google Cloud Blogs

Google Cloud has completed its acquisition of Wiz, bringing the fast-growing cloud and AI security platform into its ecosystem. The move is aimed at strengthening security for organizations running applications across hybrid and multicloud environments, especially as AI-driven systems expand. Wiz’s technology maps cloud infrastructure in real time to identify risks, detect potential attack paths, and help teams fix vulnerabilities early in development. Combined with Google Cloud’s security tools and Gemini capabilities, the partnership is intended to help enterprises detect and respond to increasingly sophisticated cyber threats.

Another Crazy Day in AI: Your Slides Look Great and That Might Be the Problem

Mar 10
3 min read

Hello, AI Enthusiasts.

Here's another crazy day in AI:

Pretty slides aren't enough anymore
OpenAI targets AI security with Promptfoo deal
Claude Code debuts agent-based Code Review
Some AI tools to try out

🎧 Listen to a quick breakdown of today’s stories.

Your Slides Look Great and That Might Be the ProblemAnother Crazy Day In AI: The Podcast

3:07

TODAY'S FEATURED ITEM: Stop Building AI Presentations That Say Nothing

Image Credit: Wowza (created with Ideogram)

When was the last time you sat through a presentation that actually said something?

Not just looked good — but made a clear argument, held up under questions, and left people with something concrete to think about. Jeff Su, a former Googler turned educator and YouTuber, put together a video that gets into exactly this. He walks through a full workflow for building presentations using Gamma, an AI presentation tool — but the bulk of his focus is on the revision process, not the initial generation. That's the part most tutorials skip over, and honestly, it's the part that determines whether a deck is actually useful or just easy on the eyes.

A few observations from the walkthrough:

The first draft usually isn’t the hardest part. Creating slides can be quick, but refining them after feedback often takes much longer.
Strong titles help carry the narrative. If the slide titles are clear enough, someone should be able to follow the overall story even without reading every line of text.
AI can help with structural edits. Moving slides, adding new sections, or reorganizing content can be done quickly with the help of AI tools.
Numbers and charts still require careful review. While AI can generate visuals or pull data, figures and sources should always be checked.
Specific instructions tend to produce better results. Clear limits or constraints can lead to more useful edits when working with AI.
Translation tools can help with global collaboration. Entire decks can be translated into another language quickly, though they often benefit from additional review.
Human judgment remains essential. Deciding what matters, how to present it, and how to respond to feedback still relies on thoughtful decision-making.

AI tools have gotten reasonably good at handling the repetitive, time-consuming parts of building a presentation. The sourcing, the restructuring, the formatting — a lot of that is faster now. And for teams managing multiple rounds of feedback across different stakeholders, that kind of speed is genuinely useful. Less time spent on the mechanical stuff means more time spent on the parts that actually matter.

The harder question is what those parts are. Knowing which feedback to act on, deciding what stays and what gets cut, understanding what an audience actually needs to walk away with versus what just sounds good on a slide — none of that gets simpler with better tools. If anything, when the tools handle more of the busywork, the quality of your thinking becomes more visible, not less. That's worth sitting with the next time you're building a deck.

Read the full blog here.

Watch it in action here.

OTHER INTERESTING AI HIGHLIGHTS:

OpenAI Targets AI Security With Promptfoo Deal

/OpenAI Newsroom

OpenAI plans to acquire Promptfoo, a tool widely used by developers to evaluate and stress-test AI systems. The platform helps identify vulnerabilities such as prompt injections, jailbreaks, and data leaks before AI applications are deployed. Once finalized, the technology will be integrated into OpenAI Frontier, strengthening built-in testing, governance, and security tools for companies building AI agents. The move reflects the growing demand for systematic evaluation and oversight as AI systems become more embedded in real business workflows.