app03 Thu, 20 Aug 2026 17:35:18 +0000 Fireside (https://fireside.fm) Software Defined Interviews - Episodes Tagged with “Agents” https://www.softwaredefinedinterviews.com/tags/agents Thu, 20 Aug 2026 09:00:00 +0200 Deep discussions about technology, enterprise IT, and the like en-us episodic Deep discussions about technology, enterprise IT, and the like Software Defined Talk Deep discussions about technology, enterprise IT, and the like yes Software Defined Talk softwaredefinedinterviews@cote.wtf Episode 125: Running D&D with AI, Determinism First, and Workflow ROI, with Michael Rishi Forrester https://www.softwaredefinedinterviews.com/125 3773701a-4bc5-418d-89fc-1f1c59419585 Thu, 20 Aug 2026 09:00:00 +0200 Software Defined Talk 125 Running D&D with AI, Determinism First, and Workflow ROI, with Michael Rishi Forrester full Software Defined Talk Michael Rishi Forrester built his son a homebrew tabletop campaign with AI in the loop - 20 hours of work instead of 60 to 100 - and found out the hard way that the model had quietly started doing the dice math itself. That turns into an argument about why individuals get so much value out of AI while organizations get so little, and why the ROI is at the workflow level or nowhere. 1:13:29 yes <p>Michael Rishi Forrester does AI workforce transformation at Accenture LearnVantage, after three years at KodeKloud and about thirty years in IT before that. Whitney and Coté talk with him about running tabletop role-playing games with AI. He built his 19-year-old son a homebrew campaign on top of Lancer, the sci-fi mech game, in about 20 hours instead of the 60 to 100 it would have taken him before - and got a website, a virtual tabletop and some Discord bots out of it. What breaks is the interesting part: there was a night when every shot did one damage, because the model had quietly stopped using the dice harness and started doing the math itself.</p> <p>From there it turns into an argument about where AI actually pays off at work. Michael’s position is that the ROI is at the workflow level and there is no organizational ROI to go find, that read-only work over more information than a person can hold is where it currently earns its keep, and that everything he has had success with is a bound system with a finite rule set. Whitney names the through-line: what can be deterministic should be deterministic, and if you need an LLM, scope it tightly and have it write the deterministic thing.</p> <p>It ends on mantis shrimp.</p> <p>You can watch <a href="https://www.youtube.com/watch?v=HRVtMHbIAR0" rel="nofollow noopener">the video version of this episode</a> as well, if you prefer that kind of thing.</p> <p>Mentions:</p> <ul> <li> <a href="https://michaelrishiforrester.com/" rel="nofollow noopener">Michael Rishi Forrester’s home page on the World Wide Web</a>.</li> <li> <a href="https://antiphonthegame.com/" rel="nofollow noopener">Antiphon: The Game</a>, the homebrew Lancer campaign - the whole story, the missions, the board, and the sea-creature capital ships.</li> <li> <a href="https://www.linkedin.com/in/michaelrishiforrester/" rel="nofollow noopener">Michael on LinkedIn</a>, where he posts most often.</li> <li> <a href="https://massifpress.com/lancer" rel="nofollow noopener">Lancer</a>, the sci-fi mech game Antiphon is built on top of. Special Guest: Michael Rishi Forrester.</li> </ul> Michael Rishi Forrester, Accenture LearnVantage, KodeKloud, AI workforce transformation, Dungeons & Dragons, Lancer, Antiphon, solo role-playing, tabletop RPG, application harness, determinism first, deterministic wrappers, MCP, Claude Code, agentic AI, hooks, LLM math, dice rolls, human in the loop, Air Force doctrine, intuition, judgment, enterprise ROI, workflow ROI, bound systems, read-only agents, context management, agent parallelism, alert fatigue, SOC 2 compliance, code review, regulated industries, layoffs, developer productivity, MIT Nanda report, Kubernetes, CI/CD, TDD, platform engineering Michael Rishi Forrester does AI workforce transformation at Accenture LearnVantage, after three years at KodeKloud and about thirty years in IT before that. Whitney and Coté talk with him about running tabletop role-playing games with AI. He built his 19-year-old son a homebrew campaign on top of Lancer, the sci-fi mech game, in about 20 hours instead of the 60 to 100 it would have taken him before - and got a website, a virtual tabletop and some Discord bots out of it. What breaks is the interesting part: there was a night when every shot did one damage, because the model had quietly stopped using the dice harness and started doing the math itself.

From there it turns into an argument about where AI actually pays off at work. Michael’s position is that the ROI is at the workflow level and there is no organizational ROI to go find, that read-only work over more information than a person can hold is where it currently earns its keep, and that everything he has had success with is a bound system with a finite rule set. Whitney names the through-line: what can be deterministic should be deterministic, and if you need an LLM, scope it tightly and have it write the deterministic thing.

It ends on mantis shrimp.

You can watch the video version of this episode as well, if you prefer that kind of thing.

Mentions:

Special Guest: Michael Rishi Forrester.

]]> Michael Rishi Forrester does AI workforce transformation at Accenture LearnVantage, after three years at KodeKloud and about thirty years in IT before that. Whitney and Coté talk with him about running tabletop role-playing games with AI. He built his 19-year-old son a homebrew campaign on top of Lancer, the sci-fi mech game, in about 20 hours instead of the 60 to 100 it would have taken him before - and got a website, a virtual tabletop and some Discord bots out of it. What breaks is the interesting part: there was a night when every shot did one damage, because the model had quietly stopped using the dice harness and started doing the math itself.

From there it turns into an argument about where AI actually pays off at work. Michael’s position is that the ROI is at the workflow level and there is no organizational ROI to go find, that read-only work over more information than a person can hold is where it currently earns its keep, and that everything he has had success with is a bound system with a finite rule set. Whitney names the through-line: what can be deterministic should be deterministic, and if you need an LLM, scope it tightly and have it write the deterministic thing.

It ends on mantis shrimp.

You can watch the video version of this episode as well, if you prefer that kind of thing.

Mentions:

Special Guest: Michael Rishi Forrester.

]]>