# Grace KY Man

Grace KY Man teaches product, growth, and marketing teams to design AI agents as connected systems, with human approval and safeguards built into the process.

- Profile: https://b2binfluencer.marketing/creators/grace-ky-man/
- LinkedIn: https://www.linkedin.com/in/grace-ky-man
- Topics: artificial-intelligence, sales, ai-agents, agentic-ai
- Based on 11 dated posts, latest 2026-09-23

## Summary

Most useful for teaching B2B teams to design connected AI agent growth systems with human gates.

- Best for: A B2B SaaS company selling agentic AI that wants a paid workshop or content series.
- Not for: Consumer or direct-to-consumer sponsored placements with follower reach targets and rate-card pricing.
- Check first: Ask for audience breakdown, sponsor policy, rate card, disclosure practice, and booking route.

### Takeaways

- Agent workflow has signal, interpretation, action, and a human gate.
- Set guardrails before the agent runs, including cost, privacy, and delegated actions.
- Trigger agents on real product events with thresholds, not timers.

## Main ideas

- **The architecture, not the model, makes agentic growth compound**: Grace KY Man argues that separate tools only add up, while one shared system compounds. In the old arrangement, a content tool, an outbound tool and a CRM do not talk to each other. The replacement is a single growth operating system with one shared store for context, content, CRM data and outbound state, which every agent reads from and writes back to. The result is a chain: research improves content, content attracts better leads, leads enrich the CRM, CRM insights sharpen outbound, and outbound conversations generate new signals. In this framing, the unit of work is the operating system, not the tool.
- **Set guardrails before the agent runs, not afterwards**: Grace KY Man treats safeguards as a precondition for running an agent and names three categories. Cost: set a per-action ceiling and model tier; once the cap is passed, the agent halts and hands off to a human. Privacy: scope available data tier by tier; whatever context happens to be nearby is not a privacy policy. Intelligence: specify which actions the agent may take alone and which require a human first. Another post states the same boundary from the operator's perspective: the agent drafts, the human decides at every gate, combining machine speed with human judgment.
- **Trigger agents on real product events, not timers**: One post sets out a four-step loop that readers can reuse as an exercise. A signal is a real product event with a threshold, such as onboarding drop-off, an approaching usage limit or a feature going flat—not something that fires on a schedule. Interpretation scores intent and routes based on ambiguity, sending clear signals to a rules engine and messy ones to a larger model. The action is tailored to the exact account and behaviour, never a template blast. The gate puts a human on anything irreversible, and the action's result becomes the next signal, so the loop learns from its own output.
- **Strategy and frameworks come before the build**: Grace KY Man sets out the order of work explicitly: strategy first, then the agents; frameworks first, then the build. AI Strategy League focuses on helping product, growth and marketing leaders architect, launch and build agentic AI systems across product growth, marketing and CRM, producing agent teams that bring in clients and content on repeat. The same sequence appears in the teaching: participants first set goals for AI projects, define workflows and establish governance, then build the agents.
- **Teaching through live builds, cohorts and reusable working documents**: Many posts show the method in action rather than arguing for it. Grace KY Man teaches through cohorts, live sessions and multi-date builds, with working artifacts including swipe-through carousels, a pricing-page example, a fill-in guardrail sheet and a personalised 90-day execution plan that participants leave with. The vocabulary stays consistent across posts and formats: agentic growth loop, growth operating system, shared store, signal, guardrail, human at the gate, and the distinction between a stack that adds and a system that compounds.

## Work to read

- [Designing the agentic approach that grows your AI product (session recap)](https://www.linkedin.com/posts/grace-ky-man_design-agentic-workflow-that-grows-your-ai-activity-7488608881431138304-3kVd)
- [Agentic AI Operating System for Product Growth and Marketing: cohort takeaways](https://www.linkedin.com/posts/grace-ky-man_in-our-most-recent-cohort-agentic-ai-operating-activity-7473374077911601153-CmaO)
- [Launching no-code growth and marketing agents in Claude](https://www.linkedin.com/posts/grace-ky-man_agenticworkflow-contentagent-outboundagent-activity-7472181439632068608-atw0)
- [AI Strategy League is expanding: senior domain experts wanted](https://www.linkedin.com/posts/grace-ky-man_agenticaiexperts-aiexperts-agenticai-activity-7474838621871738880-w5JZ)
- [Closing the loop: one growth OS instead of siloed tools](https://www.linkedin.com/posts/grace-ky-man_agenticgrowth-growthloop-agenticgrowthloop-activity-7472664329654444032-yjxN)

## Brand relationships

- **AI Strategy League** (disclosed_partnership)
- **Claude** (mention)

## Partnerships

Grace KY Man's commercial offers are teaching and build engagements rather than a published sponsorship package. The cohort course, "Agentic AI Operating System for Product Growth and Marketing," is sold by registration and covers an operating framework for agent teams, live demos of research, content, outbound, and CRM agents, and a personalized 90-day execution plan. A separate Claude-based build course was promoted across three dates in June. A one-hour live session on designing agentic workflows drew 103 attendees from more than 1,000 registrations and featured guest Godhuli Chatterjee on the product and cost side. Grace also leads AI Strategy League, describes client work as building an AI-native growth system with the client rather than handing one over, and invites senior product, growth, or marketing operators to build and deliver together. Ask Grace directly whether sponsorships are available and, if so, for rates, audience details, disclosure practice, and the booking route. Contact is invited informally via LinkedIn message or email if the sender already has the address.

## Alternatives

- Jared Waxman: Choose Jared Waxman when the need is less about teaching a growth framework to a client's team and more about the plumbing underneath: which context layer a company is building, how a governed canonical layer gets a release cycle, and how a sales org actually consumes it. Waxman runs GTM Engineer School on a folder of markdown files in a GitHub repo and brought that setup in front of Sam Gong, who standardized WorkSpan's marketing org on Claude plus a context layer, and Zach Vidibor, who keeps that layer current with Octave. He names three context layers a company can be building—brand, canonical and deterministic—and stresses that most companies say "context layer" without knowing which one they mean. He also separates builders from consumers, noting a salesperson never moves into Claude Code while the campaigns team and RevOps go in first. He argues for scoping the project to one problem with a deadline rather than "build a company brain." He offers an ongoing distribution channel: a weekly GTM Engineer Pulse issue, a Cohort 5 capstone starting 26 October where 50 students each build one real workflow for one real company, and free 30-minute calls with no promised deliverable. For a B2B SaaS buyer whose workshop should end with a working context or enrichment setup rather than an education deck, this is a closer match than a loop-and-guardrail curriculum.
- Gaurav Bhattacharya: Choose Gaurav Bhattacharya when the engagement should produce a written architecture argument rather than a live teaching series—for instance, a whitepaper or long-form piece aimed at enterprise buyers about where AI systems break inside production. Bhattacharya's framing is that most enterprise workflows are limited by orchestration, not intelligence. He lists the people acting as the orchestration layer between disconnected systems: one person checking intent signals, another enriching accounts, another prioritizing outbound, another updating CRM fields, then follow-ups, routing, sequencing, reporting and approvals. He argues the next wave is operational AI that monitors signals continuously, coordinates actions across software, escalates exceptions and operates with human oversight instead of human dependency. He proposes a stack of signals, reasoning, orchestration, action, monitoring and human oversight. He names the hard problems once an AI system is in production—orchestration, reliability, memory, permissions, governance, recovery and observability—and points readers to a digital worker stack whitepaper. He also makes the counterintuitive case that when everyone can spin up AI-powered outbound in days, execution stops being the moat and trust, positioning, distribution and signal quality matter more. For a buyer who wants a defensible written point of view on operational AI to put in front of enterprise prospects, this is a stronger fit than a workshop built around a repeatable client exercise.
- Matthew Sutherland: Choose Matthew Sutherland when the need is to reach small-business owners or operators rather than a B2B SaaS product's buyers, and when timeliness matters more than a durable framework—for a launch explainer, adoption playbook or cost-and-tier breakdown delivered while a tool release is still news. Sutherland breaks down Anthropic's Claude for Small Business release in plain structure: what the package is; that it ships inside Claude Cowork with connectors to QuickBooks, PayPal, HubSpot, Canva, Docusign, Google Workspace and Microsoft 365; that it carries 15 ready-to-run agentic workflows across finance, operations, sales, marketing, HR and customer service plus 15 reusable skills; and that every task is user-initiated, with nothing sent, posted or paid until the owner approves. He also documents the guardrails that matter to a sceptical SMB owner: existing app permissions are respected, and on Team and Enterprise plans Anthropic does not train on business data by default. The second post here is a dated cost-arbitrage play for Claude Code: the 13-day window when a new model was included in Pro, Max and Team plans before flipping to usage credits; the non-linear case for the top tier for a build sprint; the rolling five-hour reset that makes front-loading pointless; and the note that certain categories route to a different model. For a buyer who needs a fast, accurate, well-organised explainer aimed at a small-business audience, this is more directly useful than a framework workshop.

---
AI-assisted summary of public posts. Audience demographics are not available. Engagement appears only where we checked recent posts. Method: https://b2binfluencer.marketing/methodology/