❓ Problem — Stat validation

❓ Solution — LangGraph vs n8n: first build target

Metric	LangGraph	n8n
GitHub stars	~22.5k	~175k (top 0.01% of all repos)
Monthly downloads	54–55M PyPI	100M+ Docker pulls
Registered users / companies	25k+ teams on LangSmith; ~35% Fortune 500	230k+ active users; 3k+ enterprise customers
Funding / valuation	$125M Series B; $1.25B valuation (Oct 2025)	$180M Series C; $5.2B valuation (SAP investment)
Typical user	AI/ML engineer, LLM-ops, senior SWE; Python-first	Technical ops, indie hacker, back-end dev; visual-first
Enterprise fit for Workflowify	Enterprise AI/LLM-ops → strong fit	Enterprise IT automation → general fit; agent framing harder to sell

→ Build LangGraph exporter first. Hermes/OpenClaw community is Python-first and agent-native — LangGraph output is immediately useful. Enterprise LLM-ops buyers already use LangSmith (1B+ traces processed) and will see value. Build the n8n exporter in parallel at low cost; it becomes the enterprise IT expansion play.

The abstract workflow graph → dual exporter architecture means the one-or-the-other choice is an ordering decision, not an architectural fork.

Competitive landscape — trace-to-workflow space

⚠️❓ Unfair Advantage — Community size & endorsement contacts

→ Both communities are large and active. Endorsement pursuit is viable. Approach: contribute a demo workflow template to each community before v1 ships; request maintainer feedback in public rather than asking for a formal endorsement upfront.

Warning stands on moat — community adoption is a strategy, not a moat. First-mover + template library compound effect is the long game.

⚠️❓ Channels — Enterprise CAC & LinkedIn outreach

Segment	ACV range	Typical CAC	Source / confidence
SMB	$6k–15k	$300–800	First Page Sage; Averi.ai 2026 — High
Mid-market	$15k–24k	$800–1,400	SaaS Capital 2025; Benchmarkit 2026 — High
LinkedIn outreach premium	—	+25–35% vs avg	2024–2026 AI-powered acquisition data — Medium

• Healthy CAC-to-LTV ratio: 4:1 (2026 benchmark, up from historical 3:1). Early-stage: 2.5:1 acceptable during PMF validation.
• Sales cycle at this ACV: 90–180 days (compliance review, cross-functional sign-off extends vs. pure SMB at 60–120 days).
• Contract mix: 45% annual, 42% monthly, 13% multi-year; pay-as-you-grow is gaining (12% of new SaaS spend).

→ CAC modelled for Enterprise bare + options ACV ($600–6,000/yr): assume $300–800 per logo at Pro ACV; $600–1,400 at Enterprise bare. 4:1 LTV/CAC target achievable with annual contracts. Note: original $500–2k/month enterprise pricing superseded — see pricing decisions below.

✓ Decided Customer Segments — LoB-first entry strategy

Research finding below. Strategic decision (A7): LoB-first entry, not CIO-first outreach. See updated verdict.

Stakeholder	Role in purchase	Priority concern
CIO / CDO / VP AI	Economic buyer of Enterprise tier — controls infrastructure, risk, AI budget	Governance, compliance, audit trails, total cost of AI
IT / Infrastructure	Implementer and ongoing operator	Integration, security, support, self-hosting option
Line-of-business manager	Entry buyer (Pro tier) — expenses $19.90/mo independently; no IT clearance needed	Workflow delivered today; cost saved; no procurement delay

→ Strategic decision: LoB-first. Pro at $19.90/mo is below typical $50/mo corporate expense threshold — LoB buys without IT involvement (156-day procurement cycle bypassed). Push: LoB adopts and proves value. Pull: IT/CIO sees Shadow AI risk and unsanctioned usage → purchases Enterprise bare + options for governance. CIO/CDO remains the Enterprise check-signer; the entry path changed, not the ultimate buyer.

⚠️ Cost Structure — LLM API cost per workflow generation

Scenario	Calls	Input tokens	Output tokens	Claude Sonnet 4.6	GPT-4o
Low (small trace, 2 calls)	2	10k	2k	$0.06	$0.045
Mid (typical trace, 3 calls)	3	30k	6k	$0.18	$0.135
High (large trace, 4 calls)	4	60k	12k	$0.36	$0.27
Optimised (caching + batching)	3	30k	6k	$0.02–0.05

Pricing basis: Claude Sonnet 4.6 — $3.00/1M input, $15.00/1M output. Optimised scenario uses Claude prompt caching (90% input reduction) + Batch API (50% discount). GPT-4o Mini for classification steps reduces further.

• At 10 generations/user/month, Pro tier ($19.90): LLM cost = $0.40/user/month (optimised) = ~2% of revenue. Manageable margin.
• Free tier rate-limiting is essential: each uncapped free generation at mid-range = $0.18 gross API cost.
• Optimisation path (caching → batching → GPT-4o Mini for classification) can cut COGS by 85%+ vs. naïve single-model approach.

→ Free tier economics are viable with rate limiting (e.g. 5 generations/month). Prompt caching must be implemented from day one. Cost warning is resolved — model is now quantified.

✓ Decided Revenue — Pricing decisions & LTV

Decided pricing model (A6): hard caps + additional units; annual = 10 months

Tier	Monthly	Annual	Model
Free	$0	—	Hard generation cap; blocked when exceeded
Pro	$19.90	$199	Hard cap; buy additional Pro unit when surpassed
Enterprise bare	$49.90	$499	Hard cap; buy additional Enterprise unit
Enterprise + options	TBD	TBD	Modular add-ons: SSO, audit, self-host, guardrails, analytics, team sync
Web service (v2/v3)	TBD	TBD	Subscription with hard deployment cap; add units; same modular options

Competitive context (benchmark pricing)

Product	Pro / Team tier	Notes
Make (Integromat)	$9/mo (10k ops) · $16–29/mo	Low-code automation reference floor
Activepieces / Pabbly	$19/mo	Closest price match to Workflowify Pro
n8n Cloud	$50–60/mo	Self-host is free; cloud is above Workflowify
LangSmith Plus	$39/seat/mo	Per-seat; no workflow generation
Workflowify Pro (decided)	$19.90/mo · $199/yr	Below threshold for corporate expense approval; LoB self-serve

LTV estimates (updated pricing)

Tier	MRR	Monthly churn	Avg tenure	LTV estimate	CAC budget (4:1)
Pro	$19.90	3–6%	17–33 months	$340–660	$85–165
Enterprise bare (annual)	$41.58 (annual equiv.)	1–2%	50–100 months	$2,500–5,000	$625–1,250
Enterprise + options	TBD (options-dependent)	1–2%	50–100 months	Significant — options ACV determines this	Significant

Freemium-to-paid conversion benchmarks

• Developer SaaS (no credit card): 2–4%
• AI-native products: 8–12% — Workflowify targets this band with strong onboarding

Competitive landscape — closest alternatives (from pain point research)

• Lobster engine (OpenClaw community, Feb 2026) — deterministic runtime achieving 20–40%+ token reduction on routing-heavy pipelines (note: "60–80%" figure circulated in community is unverified in primary sources — see 04-02-competitive). Typed pipeline definitions. Key difference: requires manual authoring; Workflowify generates automatically from traces.
• Temporal.io — durable workflow orchestration; general-purpose; not agent-trace aware; high integration effort per project.
• LangSmith / Langfuse — observability; surface costs; do not generate workflows from traces.
• Power Automate / Zapier / Make — business process automation; not agent-trace aware.

Positioning decision (A10): "Made for AI builders"

Explicit developer-first positioning pre-empts "too complex" criticism. Target user: AI developers who know what a trace is. LoB non-technical users are served by the web service (v2/v3), not the CLI.

→ Pro LTV thin at $19.90 ($340–660) — volume and community-adoption play. Enterprise bare LTV ($2.5k–5k) is modest; Enterprise + options is the revenue engine (ACV driven by option uptake). Annual contract lock-in essential. Option pricing not yet set — key open item before Enterprise sales begin.

Validated New Pain — Skills are session-bound; autonomous scheduling requires re-platforming

Claim (user): Skills/agents built interactively on Claude, Gemini, or OpenAI cannot be scheduled for autonomous execution without either significant custom infrastructure or a switch to per-token API billing. This is a distinct pain from the implementation gap (Pain #1) and token overconsumption (Pain #2) — it surfaces after a skill has already been built and validated.

Evidence — community signal (June 2026)

Source	What it documents	Signal strength
GitHub anthropics/claude-code #28229 — "Native agent scheduling / daemon mode"	"Claude Code sessions are ephemeral — when a session ends, all background work dies. There is no native way to: run an agent on a schedule, keep an agent alive between sessions, watch a resource and wake an agent when it changes." Workaround documented: 400+ lines of cron + Python polling + tmux wake + state deduplication — custom infrastructure that every multi-agent Claude Code deployment must independently replicate.	High Official tracker
GitHub anthropics/claude-code #43397 — Cloud scheduled tasks: MCP tools not loaded	Cloud scheduled tasks cannot access MCP connectors; tools not injected into scheduled sessions. Workaround: prefix every scheduled task with "DO NOT DO ANY WORK. ASSIGN THIS TASK TO AN AGENT." so a subagent initialises the tools properly. "This effectively makes scheduled tasks useless for any workflow that depends on MCP integrations."	High 30+ confirming comments
GitHub anthropics/claude-code #47933 — Remote trigger cron never fires	Cron expressions registered via RemoteTrigger API show enabled: true and correct next_run_at, but never fire. Manual triggers work. Silent failure — no log, no notification.	High Official tracker
Paperclip (github.com/paperclip-agent) — "heartbeat scheduling" for AI agents	Open-source framework formalising the pattern: agents wake on schedule, check for work, execute with persistent context, sleep. Org charts, budgets, task queues, multi-company isolation. 43,000 GitHub stars in under 6 weeks — strongest community demand signal found.	High Exceptional organic growth
Anthropic billing reclassification — June 15, 2026	Automated usage (claude -p, Agent SDK, GitHub Actions) moved to a separate monthly credit pool at API list rates. Interactive use (chat, Claude Code terminal, Cowork) stays on subscription. Pro credit: $20/month — consumed in a single day by 5 parallel agents. "On the other side sits everything a script or a scheduled job does for you while you are not watching. That side now pulls from the new monthly credit." Official Anthropic change.	High Official Anthropic policy
CodeGeass (github.com/DonTizi/CodeGeass)	Third-party tool built specifically to schedule Claude Code agents via cron using subscription tokens. Features: CRON expressions, multi-project management, skills system, plan approval gates, Telegram/Discord notifications. Product category exists because the platform doesn't solve this natively.	High Market validated by product existence
Community practitioner posts (dev.to, GTM Architects, dortort.com, kjetilfuras.com)	Multiple independent writeups on running Claude Code agents as 24/7 daemons. Each documents the same infrastructure required: process manager (pm2 / launchd / systemd), state files, lock files, retry logic, MCP workarounds, context window management. Boucle (autonomous loop framework, MIT-licensed) emerged from this need. Estimated engineering effort to get from "it works in a session" to "it runs on a schedule reliably": 20–40 hours.	High Multiple independent sources

Why this is a distinct pain (not a sub-point of Pain #1)

Pain	When it surfaces	What the user has already done	What blocks them
Pain #1 — implementation gap	Before any workflow exists	Observed a workflow work once in a session	Dev time + IT clearance + weeks to build a production workflow
Pain #4 — scheduling gap (new)	After a skill/workflow has been built and validated	Built and tested the skill; it works interactively	Platform is session-based; scheduling requires API billing + 20–40h custom infrastructure

The "scheduling gap" is the next wall after the implementation gap. A developer who used Workflowify to generate a LangGraph workflow hits it immediately: "I have the workflow — now how do I run it every morning without opening a Claude session?"

Workflowify's answer to this pain

• Generated workflows are not session-bound. A LangGraph Python script or n8n JSON workflow runs as a standalone process — no Claude session required for execution.
• Native scheduling support is a first-class v1 output: wire up via launchd (macOS), systemd timer (Linux), n8n scheduler, or LangGraph cron — one CLI flag.
• No subscription quota consumed at run time. Deterministic nodes are pure Python/API calls; only LLM nodes consume tokens, and those use cheap models (Haiku / GPT-4o Mini), not Claude subscription quota.
• Cost model at scale: a deterministic LangGraph workflow running nightly costs ~$0.001–0.01/run vs. $0.50–5.00 for an equivalent skill run via Claude API. 50–500× cost reduction for scheduled, recurring tasks.

Canvas implications

Cell	Update
Problem	Add Pain #4: "Skills and agents are session-bound — scheduling autonomous execution requires switching to per-token API billing or building 20–40h of custom scheduling infrastructure. Validated: Paperclip (43k stars in 6 weeks); Anthropic June 15 billing split; GitHub #28229."
Solution (③ Local deployment)	Strengthen scheduling as a first-class output: "Schedule via launchd / systemd / n8n scheduler / LangGraph cron — one CLI flag. No Claude session required to execute."
UVP	Consider adding: "…and schedule it — no session, no subscription quota."
Customer Segments / Early Adopters	The scheduling pain surfaces in power users who already have working skills — they are already in the target early adopter segment. No new segment needed; this deepens the existing one.

→ Confirmed: distinct Pain #4. Session ephemerality + API billing model + scheduling infrastructure complexity are three independent frictions compounding the same problem. The Anthropic billing reclassification of June 15, 2026 and the Paperclip 43k-star signal are primary evidence. Add to Problem cell. Strengthen scheduling copy in Solution and UVP.

Sources: github.com/anthropics/claude-code issues #28229, #43397, #47933 · techtimes.com/articles/317625 (Anthropic billing reclassification) · genaiunplugged.substack.com · github.com/DonTizi/CodeGeass · dortort.com/posts/closing-the-automation-gap · kjetilfuras.com/claude-code-agents-production · grow.thegtmarchitects.com · dev.to/boucle2026. Confidence: High — multiple independent sources, official Anthropic policy.