Manifesto

System design tools
for AI and humans.

labsly builds Mac-native software for the people who design how things work — systems, strategies, services, products. Each tool treats both human practitioners and AI agents as first-class users. We bet on structured input, local AI, and epistemic honesty about what AI knows and doesn't.

The category

Most software in this space is one of two things. Free-form whiteboards (Miro, FigJam, Notion canvases) — great for sketching, weak on rigor. Or thirty-year-old enterprise tools (Vensim, Stella, AnyLogic) — rigorous, but UX-hostile and expensive. Both groups treat "AI" as either marketing dust or an explicitly rejected risk.

There's a third option that hasn't existed: a structured-by-default tool with AI built in honestly at designer-quality craft. That's the category labsly builds. We call it system design tools because the underlying primitives — stakeholders, structures, simulations, critique — are what designers across every discipline actually share, even when their domains differ.

The intellectual lineage

We didn't invent any of this. labsly stands on a tradition of systems thinkers — Donella Meadows, Jay Forrester, Stafford Beer, Russell Ackoff, Peter Senge — who spent decades arguing that the way to fix complex problems is to model them, not to react to events. Their tools — causal loop diagrams, the iceberg model, system archetypes, leverage points — are field-standard vocabulary for anyone who works in strategy, public policy, organizational design, or large-scale product. They're also locked in books and PDFs and a handful of aging desktop apps.

Adjacent traditions inform other parts of the labsly Suite. Simon Wardley's mapping work for strategy. Lynn Shostack and the IDEO/Live|Work tradition for service design. Donald Norman, Marty Cagan, Teresa Torres for product design. The systems-thinking critique tradition — Peter Checkland's Soft Systems Methodology, Werner Ulrich's Critical Systems Heuristics — frames how we approach the harder boundary questions.

We acknowledge this lineage in detail on each app's Credits page. We use the field-standard names where they're field-standard vocabulary; we paraphrase where the originator's exact text is theirs to keep.

The thesis

1. Structured input beats free text.

Most AI tools try to derive structure from text. It's a lossy game — every input has to be re-parsed, every analysis is grammatical rather than architectural, every iteration is a fresh hallucination. We invert this. The user provides the structure (graphs, manifests, icebergs, personas, blueprints) and AI reasons over it. The result: critique that engages with the model you're actually building, not with whatever the AI inferred you might have meant.

2. AI agents are first-class users.

Every labsly tool exposes its capabilities as an MCP server alongside its UI. Humans use the apps directly; AI agents (Claude Code, Cursor, Cline, Aider) call the same tools via protocol. This isn't a side feature — it's a design assumption that shapes everything from API surface to error handling. The Mac app is the revenue surface; the MCP server is a free distribution channel. Both audiences are first-class.

3. Local AI is the differentiator.

Apple-Silicon-native inference is good enough to do real work without sending a single byte to the cloud. We standardize on local where it serves you, and let cloud in (Claude, OpenAI) where it adds enough value to justify the trade-off. You bring your own keys; nothing routes through us. Privacy, predictable cost, no rate limits, fully offline, fine-tunable on your taste. It also means our apps work in places — airplanes, secure facilities, slow-internet environments — where cloud-only tools don't.

4. Designer-quality Mac-native craft.

We optimize for the people who notice. Two visual tracks: a tools track (Mac-native restraint, system font, dark-mode-first, lots of whitespace — for Systema, Witness, Atlas, Dais) and a craft track (warm typography, paper textures, light-mode-friendly — for Cairn and future creative apps). Both are held to Apple-quality polish. Free alternatives exist; what labsly sells is craft.

Epistemic honesty (the moat)

December 2025 brought a turning point in the AI-personas conversation: ACM Interactions published The Synthetic Persona Fallacy, OpenReview hosted multiple papers showing LLM personas systematically distort subgroup opinion, and CHI '25 surfaced research showing persona prompting often degrades rather than improves fidelity. The market has discovered that a confident-sounding AI persona is not the same as understanding a real human.

This is the moat we're building. Every labsly tool that surfaces AI-generated content surfaces confidence, provenance (where the answer came from), and source type (stated / inferred / synthesized / observed). Personas come with confidence labels per field. CLD critiques cite the model's reasoning. Strategic counterfactuals show their assumptions explicitly. We tell you what the AI doesn't know, instead of pretending.

This is operationalized at the data-model layer (Systema's EpistemicTypes.swift: provenance, claim type, review state, slot disposition) and inherited across the Suite. It's not a feature; it's substrate.

The Suite

Four design disciplines, four tools that compose into end-to-end workflows. An AI agent — or a human — can start at any layer and propagate changes through the others. The artifacts that flow between them (personas, system maps, journey blueprints, design critiques) are open JSON formats published at labsly.com/schemas.

Adjacent to the Suite, we publish other Mac apps under the labsly umbrella that share substrate but stand on their own positioning: Dais (strategic communication coaching, Q1 2026 — first to ship), Cairn (literary craft, Q2 2026), VoidRepublic (political simulation game), Meridian (personal portfolio analytics).

The unfashionable claim

The contemporary AI-tool market sells velocity. Generate prose. Generate slides. Generate code. Faster, faster, faster. labsly sells the opposite: a tool that defaults to challenge, not validation, that asks where the holes in your model are, that surfaces what it doesn't know.

The intuition behind this: the people who design systems, services, products, and strategies are not in a velocity bottleneck. They're in a clarity bottleneck. The constraint is being honest about what you understand, what you've assumed, and what you've ignored. Faster generation doesn't help when the next step is "interrogate your own assumptions."

This is who labsly is for. Independent strategists. Senior PMs allergic to tool-stack pitches. Design directors at mid-size companies. Indie consultants. Think-tank fellows. PhD students. The people whose job is to think clearly about hard problems.

How we build

labsly is built by one person — using AI as the multiplier that makes a 6-app portfolio feasible. The build is happening in public on Substack and X. Every architectural decision, every misstep, every shipped feature gets written about. We're the proof case for the thesis: AI and humans, individually or together, building things that no one team alone could build.

It also keeps the products honest. The person designing labsly's tools has to use AI to ship, which means we know exactly where AI-augmented work helps and where it hurts. We bake those lessons into the tools.

What's next

Dais ships Q1 2026 and proves the labsly approach to one app. Cairn follows in Q2. Systema is the flagship Q3. Witness lands Q4 with both the MCP server and the Mac annotation studio. Atlas opens 2027. The Service Design app rounds out the Suite in mid-to-late 2027. labsly Pro bundles them.

If any of this resonates — if you have spent too many hours hand-pushing sticky notes around an infinite Miro canvas, or wrestled with Vensim's UX, or felt the dishonesty of an AI persona that sounded too confident — the email list below is the way to follow along. Build-in-public posts go out roughly weekly. Launch announcements when each app ships.

Occasional updates on labsly app launches. No spam, no tracking pixels. Unsubscribe anytime.

This manifesto will evolve as the Suite ships. Last updated 2026-04-22.