iiinova — where your digital self keeps evolving

Mitchel Pigsley @mitchelpigsley.art · 14/08/2026

For what I consider an "AI-first" technology company, realmsofshod.com has a ridiculous amount of non-AI-related infrastructure and systems. I suspect most "AI-first" businesses can relate.

Realms of Shod empowers tabletop gamers and storytellers to capture every moment of each adventure. Full-session recall provides automated transcripts and searchable archives, so you'll never scramble...

020

Six years, five layers, each built on the last. The sixth layer isn't about removing humans or keeping them. It's about deciding where the lights go. Humans decide. Agents execute. The future isn't lights out. It's knowing which lights to leave on.

000

Here's what I believe to be the shape of the future. Humans meet, argue, and ultimately decide. The moment a decision lands, agents take it and execute. Not just the routine work. The work. Decisions are made, and businesses shift. The decision room and the factory floor work together.

100

The dark factory injects human judgment once, at the spec boundary. A habitat injects it continuously. Where taste and judgment are the differentiator, especially humans selling to humans with human needs, that work is habitat territory.

100

I've seen a new term circulating called habitat engineering, where environments are designed for humans and agents to work together. Shared state both can read and write. Legible handoffs. Trust boundaries that move as confidence grows. Attribution baked in. Lights on, by design.

100

The entire progression rests on the assumption that the human is friction to be removed. But what if the human isn't the bottleneck these systems keep refactoring out? What if human judgment is the one input you can't spec in advance?

100

Which brings us to the dark software factory. Specs go in. Software comes out. Agents write, test, and ship the code. No human reviews the diffs. This isn't speculative. Systems exist today to adversarially review and grade code changes. Working software with no humans in the build loop.

100

Loop engineering (late 2025): the agent's work cycle becomes the design surface. Observe, act, evaluate, repeat. Graph engineering (now): subagents, parallelized work, and long-term planning. Models now run for hours through complex work. Humans became a checkpoint.

100

Harness engineering (2025): the models got good enough at multi-step work that the environment became the bottleneck. Tools, sandboxes, verification, guardrails. The uncomfortable truth of this era was that a mediocre model in a great harness beats a great model in a bad one.

100

Context engineering (2024): AI began to think before answering and to work on tasks for minutes, not seconds. A new problem arose. Thinking AI fills its own context window (its working memory) fast. And a stuffed window made it dumber. Wording mattered less. Curation mattered more.

100

Prompt engineering (2020): you type one message, the AI gives one answer. No tools, no memory, no thinking it over. This is the era that gave us ChatGPT. The only thing you controlled was the wording. So people obsessed over it as if they were uncovering magic phrases or lucky incantations.

100

Yes, some of this is buzzword churn. But each new name marks a real event, and the next bottleneck is uncovered. Each stage absorbs the layer below and builds on it. Five layers in six years, and the pace is picking up. The point is that it's moving, so let's review each prior stage.

100

There's a term for where AI software development is headed: the dark software factory. Like a lights-out warehouse, but the product is code. The path there already has five completed stages: Prompt → Context → Harness → Loop → Graph But lights-out isn't the only place this road goes. 🧵

With AI agents these days, there's power in being an individual contributor (IC). Your leverage comes from knowing the underlying systems, standards, and best practices in your area of expertise.

000

I keep thinking about the last session of Brindlewood Bay. The concept of dynamically building up a mystery into a surprising conclusion was one of the best #ttrpg experiences I've had in recent memory.

020

Become a better #ttrpg player by exploring your character's backstory on realmsofshod.com. See your sessions, messages, and backstory explorations in the ongoing gamestream.

New gamestream with backstory exploration, sessions, and scenes mixed together.

000

#ttrpg campaigns die from scheduling conflicts and lost context. Stories become forgotten. At realmsofshod.com, group world-building, multiple game master support, character backstory exploration, and live chat help keep you involved between sessions. Sign up today to claim your free coins 💰

Mitchel Pigsley @mitchelpigsley.art · 04/08/2026

100

Internally at realmsofshod.com we are calling realm agents “gardeners”. A realm is a garden where humans tell stories and the system records and surfaces it. We are excited to share more about this in the upcoming months.

Mitchel Pigsley @mitchelpigsley.art · 04/08/2026

Habitat engineering. Building spaces where agents and humans can coexist and collaborate effectively. What habitats are you building? #saas #startup #ai #agents

130

Oh yeah, I run my fare share of online games. And agree. There’s something so special about IRL play.

Listen, agentic development is a wonder, but at the end of the day it’s a tool in the toolkit. It doesn’t mean you don’t get to pay attention to security, scalability, and codebase hygiene. We are engineers, let’s act like it.

000

Feeling some intense FOMO with all the Gen Con posts. Next year I’m making it happen.

000

Covid really changed the game for virtual tabletops. A lot of people started playing #ttrpgs digitally. Now it’s all they know.

100

One the roadblocks we’ve experienced building prosumer AI tools is group consent. If only one person in a group takes a hardline stance against the use of AI, then the group abandons the platform.

000

#saas distribution means something different for every business. For example… prosumer, low-ACV applications aren’t going to be hiring a SDR or performing outbound sales. These businesses must be self-serve and often need built-in distribution to survive.

Mitchel Pigsley @mitchelpigsley.art · 02/08/2026

Okay, I dug in because I like your style. Link checks out on my end, but Bluesky shortens the display text, so if it got copied instead of clicked, that'd 404 for sure.

Mitchel Pigsley @mitchelpigsley.art · 01/08/2026

Completely agree! I shoved them into a list when there’s definitely a hierarchy.

100

Mitchel Pigsley @mitchelpigsley.art · 01/08/2026

Hah, nice. You sure it’s dead?

Mitchel Pigsley @mitchelpigsley.art · 01/08/2026

Complexity overhead quietly kills #SaaS startups. We deliberately deferred ours to fight the dragons that actually matter: bad distribution, weak retention, building something nobody wants. Read the story: realmsofshod.com/blog/complex... #buildinpublic #indiehackers #startup #webdev

Complexity, Deliberately Deferred | Realms of Shod Blog

Starting a technology company, and specifically a B2C SaaS company, has its fair share of challenges. I'm sure this comes as no surprise. The technology is one…

380

Mitchel Pigsley @mitchelpigsley.art · 20/07/2026

I've always been interested in the history of #ttrpgs, which sent me down the Kriegsspiel rabbit hole. This essay traces that lineage from the earliest versions of Dungeons and Dragons to modern online play, and explains why we so often reach for VTTs. realmsofshod.com/blog/wargami...

Wargaming and the Proliferation of VTTs | Realms of Shod Blog

Picture it. It’s your very first time hosting an online TTRPG game night for friends. You’re confused over how to run a game online, so you turn to the…

Mitchel Pigsley @mitchelpigsley.art · 08/07/2026

As NPCs get mentioned in a #ttrpg story, the richer they become. They change, but do your notes ever get updated? At realmsofshod.com, we’ve been working on a quick action that auto-syncs those entity descriptions in the compendium. Try it for yourself, now!

Mitchel Pigsley @mitchelpigsley.art · 05/07/2026

020

Wherever and however you play #ttrpgs with friends, realmsofshod.com has you covered. Transcriptions, scene-splitting, and compendium extraction are now available.

Host your sessions on Realms of Shod, play on Discord by inviting a bot to your voice channels, or upload audio from a phone recording, Zoom, or anything. Realms of Shod extracts your scenes and automatically builds your compendium.

011

Mitchel Pigsley @mitchelpigsley.art · 05/07/2026

#ai offers a poor facsimile of intelligence. Do I think #agi is right around the corner? No. A poor facsimile can still be useful, though.

000

Mitchel Pigsley @mitchelpigsley.art · 05/07/2026

I've given my players access to such powerful Mörk Börg equipment that I have to throw enemy numbers at them now to make a dent. Even the largest and most difficult of written enemies can't compete. Anyone else dealt with higher-level #morkborg characters?

Mitchel Pigsley @mitchelpigsley.art · 01/07/2026

#AI has a marketing problem. Anthropic and OpenAI have evangelized LLMs as a tool to replace human labor. In addition to the resources these data centers consume, it's no wonder they are widely hated. When the hype cycle dies down, I hope we can clearly see the potential of these tools.

100

Mitchel Pigsley @mitchelpigsley.art · 16/06/2026

I can't wait to continue my first playthrough of Disco Elysium.

000

Mitchel Pigsley @mitchelpigsley.art · 16/06/2026

A #storytelling MCP that fits right into your #ttrpg live play. Use agentic AI to generate or transfer your collected notes into a coherent compendium which links itself to your ongoing session transcripts. Find out more: realmsofshod.com/docs/integra...

Connect your AI assistant to Realms of Shod using the Model Context Protocol. Once connected, your assistant can read and manage your realms, sessions, and compendium directly. No copy-pasting needed.

MCP Integration | Docs

Isolation's useful for the loud failures, I agree. The ones I get stuck on are quieter, like a missed auth check, or a query that's fine in dev and dies at scale. That stuff compounds. Decisions become harder to walk back. To me, it's a context gap more than a scope issue.

000

I'll go ahead and say it. I really like using #agentic #ai to help me flesh out and build campaigns for #ttrpgs. Claude Code running within an Obsidian Vault is actually so helpful. And if you're using tools like realmsofshod.com to transcribe your sessions, it becomes even more powerful.

022

As humans move up the intelligence stack, more delegation to #agents must occur. Agents that spawn agents that spawn agents. But I have yet to see this work in practice. I'm waiting for the watershed moment I experienced with #ClaudeCode, but at the level of multi-agent orchestration.

000

Realms of Shod @realmsofshod.com · 07/05/2026

I'm still finding that #multi-agent orchestration suffers from imprecision and small mistakes that compound negatively over time. When it comes to writing scalable, production-grade software, human intuition is still king... for now.

200

Reposted by Mitchel Pigsley

A new onboarding experience for Realms of Shod just dropped. This spring we've been rebuilding our world graph infrastructure. What's coming is worth the wait. Stay tuned. #worldbuilding #TTRPG

021

Mitchel Pigsley @mitchelpigsley.art · 03/03/2026

Claude code is taking over. I walk through coffee shops, and it’s open on every other computer.

Mitchel Pigsley @mitchelpigsley.art · 24/02/2026

I'd love to connect with readers on #storygraph. I read sci-fi/fantasy, TTRPGs, and non-fiction as well. “mpigsley” is the username. Let's connect!

Mitchel Pigsley @mitchelpigsley.art · 17/02/2026

Your #TTRPG world builds itself while you play. Realms of Shod now auto-links characters, places & things in your live transcript, and tracks their relationships. That nasty little witch will always live in that strange little cottage. Try for free @ realmsofshod.com

Aut-linked entities in a live transcript on Realms of Shod

Your realm's compendium assembles itself as you play.

030

Mitchel Pigsley @mitchelpigsley.art · 02/02/2026

I'm late to the game, but Claude Code with GitHub and Linear MCP is a cheat code. I also recently found github.com/kbwo/ccmanager, which helps with multi-agent workflows. I'm also not saying this is the endgame. It still feels like wrangling a bunch of junior developers.