Impertinent: All Things AI

The Search for Intelligence: Why QMD might be an Architectural Keel for the Agentic Era

Bill french — Sun, 08 Mar 2026 16:14:36 GMT

I recently followed a GPS that insisted on shaving exactly 120 seconds off my ETA by routing me through a primary school zone at 3:00 PM. On paper—within the “world model” of the mapping software—it was a legal, paved road with a 20 mph limit. In reality, it was a tactical nightmare of double-parked SUVs, a phalanx of strollers, and crossing guards with…

The Fatigue of the All-In-One Doc

Bill french — Tue, 27 Jan 2026 14:20:26 GMT

The UI is the Friction: Eviction of the Lego™ Apps is Intevitable

There is a specific kind of exhaustion reserved for the productivity enthusiast. It isn’t the tiredness of work done; it is the tiredness of work organized. I experience it often. I’m sure you know exactly what I’m talking about.

From Apps to Conversations - Thanks NanoBanana!

For the last d…

The Revenge of the Elders: How AI Turns "Aging" into "Ascending"

Bill french — Sun, 25 Jan 2026 14:27:29 GMT

Let’s dispense with the polite fiction that the “aging workforce” is merely coping with the modern tech stack. We aren’t coping. We are waiting for the rest of you to catch up.

Katelin Cruse recently highlighted the friction older workers face in a youth-obsessed industry. It’s a valid observation, but it misses the tectonic shift happening beneath our f…

AI Vision Changes Everything

Bill french — Sat, 17 Jan 2026 20:24:19 GMT

I’ve spent the better part of the last two years trusting a computer to hurl me down the interstate at 85 miles per hour. I drive a Tesla Model Y, and since early 2024, I’ve logged over 35,000 miles on Full Self-Driving (FSD). I use it 99% of the time.

When I first started, the skepticism was palpable. My wife, sitting in the passenger seat, was a coiled…

The Ideation Headwater

Bill french — Thu, 01 Jan 2026 17:17:30 GMT

In the sanitized world of software development, we like to pretend that “ideation” happens at a whiteboard, fueled by expensive espresso and the calm hum of a MacBook Pro. We treat ideas like scheduled deliverables—items to be logged in Jira or neatly bulleted in a Notion page during a synchronous Zoom call.

We are lying to ourselves.

The 95% AI Failure Rate: Why a Better Cage Won’t Fix Your AI Parrot

Bill french — Thu, 04 Dec 2025 19:36:12 GMT

The “95% of AI pilots fail” statistic from MIT has become the industry’s favorite haunting campfire story. It’s the specter looming over every QBR and roadmap meeting. So, naturally, when AWS takes the stage to announce Agent Core, the industry holds its collective breath.

I’ve reviewed the architecture. It’s impressive. They’ve rol…

Cascade: A Superhuman Agent for AI's Critical Blind Spot

Bill french — Wed, 05 Nov 2025 23:30:04 GMT

We are drowning in brilliant, often useless noise.

Gemini/NanoBanana

In our pursuit of artificial general intelligence, we’ve successfully developed engines that produce breathtaking insights. Tools like Claude, Grok, and their ilk can synthesize, analyze, and sometimes hallucinate at superhuman speeds. They hand us novel strategies, deep-cut data correla…

My Productivity Sandwich Just Became the Daily Special

Bill french — Wed, 29 Oct 2025 13:25:20 GMT

I’ve been eating the same lunch for years.

It’s not a sad desk salad. It’s a three-part productivity stack I call the “Grammarly Coda Superhuman sandwich.” It’s sort of like a triple-decker club.

Gemini: NanoBanana

Coda provides most of the protein in this sandwich. I think we can all agree that, in vast measure, our knowledge work depends on information—t…

Google’s “Opal” Isn’t Killing the Glue Factories

Bill french — Tue, 28 Oct 2025 13:45:19 GMT

The breathless X posts and “AI Insider” newsletters proclaiming Google’s new Opal tool as the “Zapier-killer” are analytically lazy. It’s a classic case of tech-hype myopia: seeing a shiny new object from a Big Tech company and immediately assuming it must replace every adjacent, established tool.

This is a fundamental category error. Comparing Opal to t…

The Claude Skills Dilemma: Redundant Gimmick or... a Dumb Wrapper for a $100B Idea?

Bill french — Sat, 25 Oct 2025 11:35:17 GMT

Anthropic just did that thing tech companies do: they dropped a shiny new feature on a Thursday and set the developer world on fire. On October 16, 2025, they unveiled Claude Skills. The pitch? A dead-simple way to teach Claude new tricks.

The community buzz is palpable. “It’s Claude.md++,” they say. “Lightweight.” “Agent-friendly.”

The idea is elegant. A…

VoiceLink, Typeless, and Monologue: Exploring the Future of Voice Transcription Tools

Bill french — Fri, 08 Aug 2025 22:58:00 GMT

Voice transcription is a promising technology that many expect will revolutionize productivity by capturing ideas effortlessly and transforming speech into written insight. However, as I have explored three leading competitors—VoiceLink, Typeless, and Monologue—and reflected on cutting-edge research, it’s clear that voice-to-text is far from a solved pr…

Voice: The Turning Point

Bill french — Sat, 02 Aug 2025 13:46:18 GMT

It was 40 years ago, maybe 30. I can't remember, but I started using Dragon NaturallySpeaking. Today I'm using Typeless (typeless.com). This entire article was written by voice.

For voice productivity to be compelling, it must blend naturally with the hardware device. I think that's what's going on under the hood with Typeless. They have essentially crea…

Why Your Team Should Leave Slack for Den

Bill french — Fri, 25 Jul 2025 17:29:09 GMT

For years, Slack has been the default hub for team communication. But as the demands of modern work evolve—especially with the advent of agentic AI and the need for seamless automation—forward-thinking teams are seeking more than just chat.

Enter Den: an AI-powered workspace where humans and agents collaborate, automate, and organize their work more inte…

The Context Paradox: Why Generative AI Needs More Than Just Data

Bill french — Tue, 01 Jul 2025 13:06:31 GMT

Let’s be clear: generative AI isn’t failing because the models are too simple, or because the code is buggy, or even because the underlying math is misunderstood. It’s failing—again and again—because we hand it a jigsaw puzzle with half the pieces missing and then marvel at the nonsense it produces. This isn’t a technical shortcoming; it’s a context cri…

Limitless: Deeply Limited

Bill french — Wed, 04 Jun 2025 17:04:28 GMT

I have written about Pendant, the Limitless Rewind sequel. I still love this device, notwithstanding some trepidation.

Here’s a story that runs in parallel to my feelings of uncertainty about Pendant.

In 2019, I became irrationally hyper-aligned with Coda. It was a new and delightful platform for building smart documents that possess app-like qualities. B…

I Don't Code, I Context

Bill french — Sat, 24 May 2025 21:42:24 GMT

To “code” (verb) is to create a set of instructions (the “code,” noun) that a computer executes. Coding involves structuring logic to solve problems or build systems. Agents, and generative AI LLMs for that matter perform well when we give them detailed contexts. “Contexting” (a coined verb) in the context of AI agents, actively shapes or provides the c…

Memory

Bill french — Sat, 17 May 2025 17:37:32 GMT

Since the advent of generative AI, I have reflected extensively on the concept of memory. This article was written during a time when the world was buzzing over a basic memory announcement from OpenAI, but there are plenty of other articles—just search for "memory" along with the term "Impertinent."

Memory is crucial, but poor, fragmented, or incomplete …

Choosing the Best Voice AI Platform: A Probabilistic Analysis of Deepgram, OpenAI, and ElevenLabs

Bill french — Sat, 03 May 2025 09:48:10 GMT

Building a voice AI agent is a high-stakes decision. With platforms like Deepgram, OpenAI, and ElevenLabs vying for dominance, how do you choose? Drawing on Annie Duke’s Thinking in Bets, we assess these platforms under uncertainty, assigning probabilities of success based on technical performance, customization, cost, scalability, and developer experie…

NLP is Blind: Why AI Transcripts are Dead

Bill french — Wed, 09 Apr 2025 14:00:46 GMT

StableDiffusionXL: Why Transcripts Are Dead

Let’s cut the fluff: NLP is a blindfolded butcher hacking at audio’s carcass, serving up transcripts as if they’re the whole meal. They’re not. Audio’s a living stream—pitch bending with love, volume surging with rage, silences thick with doubt—and NLP has no eyes for it. It’s like reverse osmosis water: pure, …

Record Your World with Pendant®

Bill french — Sat, 05 Apr 2025 18:45:52 GMT

I got my Pendant® a few weeks ago. The device is well-made, solidly constructed, and very reliable at a hardware level—the Limitless software that goes with it—not so much.

The Slack forum for Pendant users demonstrates what a shitshow1 it has been. The software for this wearable device wasn’t ready when this promising wearable hit your mailbox. It’s sti…