AI Agents Seattle August 2026

Adam Jacob reduced one code review from 4.6 million tokens across 23 agents to 506,000 across 3 by replacing the LLM coordinator with deterministic code. Casey Margell spent nine months facing numerous challenges in production and concluded that the biggest risk with AI coding agents is efficiently constructing the wrong product. One is about building a factory; the other is survival in the field. On August 11, they will compare notes.AI Agents Seattle is a meetup for engineers actively working with AI agents. It exists because the field is evolving faster than anyone's understanding, and the individuals involved need to share hard-earned lessons instead of quick opinions. This August, two speakers tackling the same issue from opposing starting points—one constructing an agentic factory from the ground up, the other adapting agents into an existing business—will discuss their choices, their failures, and what changes they plan next.How to build a software factory.Adam Jacob, CEO and Co-Founder, Swamp ClubAdam Jacob builds agentic systems as reproducible, auditable software factories, and his agents are producing what he describes as remarkably consistent architecture. He will share how he streamlined a complex code review load from 4.6 million tokens across 23 agents to 506,000 tokens across 3 by shifting coordination out of the LLM and into deterministic code, and why he believes it's premature to invest in frameworks when adaptive building blocks are prevailing.Adventures in Adapting: nine months facing challenges building AI agents. Two very different approaches, the compromises I made, and how I assess what's next in a realm that continues to evolve.Casey Margell, Co-Founder and President/COO, Indisea SoftwareCasey Margell has spent nine months developing AI agents in production, most recently integrating Claude into Indisea's Slack and connecting it to Jira and Confluence; it filed 9 correctly formatted Jira tickets from a PDF of meeting notes on the first day. He will discuss why he believes the greatest risk with AI coding agents is efficiently building the incorrect product, and the product-owner-agent model he employs to keep agents aligned with requirements after he steps away.