Howdy 👋🏾.

This week, I had the pleasure of joining an amazing panel at DC Startup and Tech Week to talk about agentic AI and automation. While preparing, I kept staring at the word automation and thinking it was wrong. It somehow misses the point of the moment we’re in, to the extent that I’m not sure we should even use that word when talking about AI.

Here’s the thing: automation is about building something repeatable. It’s the dream of the perfect process. You define the steps once, and the system does them over and over again, the same way, every time. Think of an assembly line, a payroll run, or an email sequence that always fires in the same order. That’s automation.

AI is different. What AI offers is closer to augmentation. It adds judgment and adaptability to systems in ways that weren’t possible before. AI is non-deterministic, which is a technical way of saying that, like a human, it won’t do the same thing twice. If I ask ChatGPT a question and you ask the same one, we’ll get different responses. That’s by design. Inside a system, it means AI can follow a process and read the room to consider context, weigh tradeoffs, and change the script when it makes sense.

At PerryLabs, we avoid the word automation and talk instead about workflows. Workflows let us use tried-and-true automation where consistency matters, and bring AI in where judgment matters.

A concrete example: on one client project, we’re automating the parts that should never be subjective, like checking page margins, validating file formats, confirming font sizes, and required fields. No AI needed. Then we use AI for the parts that live in the gray. Does this piece of writing feel on-brand for the audience? Is this image balanced for the layout and goal? Should we escalate this submission because something looks off, even though it passes the rules? That’s not automation. That’s intelligence.

This is where AI becomes really powerful: the messy middle. The work is too nuanced for simple rules. A huge swath of knowledge work has been stuck there, awkward, half-automated, forever “needs a human to check.” That’s exactly where AI can make the biggest impact.

-jason

🎙️Fresh From the Studio

In Episode 1 of Thoughts on Tech & Things, I talk with my friend Ben Slavin about OpenAI’s DevDay, Sora 2, ChatGPT 5, and how tools like AgentKit and the App SDK are shifting OpenAI from product to platform.

Listen now and subscribe on Apple or Spotify. If you enjoy it or have ideas for future episodes or guests, I’d love to hear from you. I read every message you send, just hit reply or email me at contact@jasonmperry.com.

🔗 The Best in Tech This Week

🧭 OpenAI Launches the Atlas Browser – OpenAI just dropped its own AI-powered browser, Atlas, joining Perplexity’s Comet in aiming for Chrome’s dominance. These new AI browsers blur the line between search, chat, and productivity, making the address bar into a conversation.

📜 Reddit Sues Perplexity for Content Scraping – Reddit is suing Perplexity for allegedly pulling Reddit content into its AI search results. It’s a big test for what “open web” really means in the age of LLMs, especially as AI browsers and search tools find creative ways to get around robots.txt and Cloudflare barriers.

📎 Clippy’s Back with an AI Upgrade – Microsoft’s new “Mico Copilot” brings Clippy back from the dead. It’s smarter, faster, and far less annoying… hopefully. The company swears this is the helpful version we always wanted in 2001.

🎤 The AI Roadshow: Workshops, Talks & Beyond

Oct. 28 – Enoch Pratt Library Event on XR/AI
Oct. 30 – Leadership in the Age of AI
Nov. 3 – Greater Baltimore Climate Summit
Nov. 12 – WTCI AGILE Series
Nov. 20 – AI Summit Europe

P.S. Before you go…

An AWS outage took down half the internet this week, and one of the strangest casualties was the Sleep8 smart bed. When AWS’s DNS went offline, the beds couldn’t reach their cloud endpoints, so heaters stayed on, and automatic adjustments stopped working. A wild reminder that even your mattress might depend on Amazon’s uptime.