Human intent
Goals, brand rules, commercial constraints, approvals, and ownership.
Agentic ecommerce systems & growth operations
I build the operating system around ecommerce agents.
I connect client isolation, reusable brand skills, Slack workflows, Shopify, Klaviyo, paid-media and CRO operations, KPI receipts, and human approvals into a production system a growth team can inspect, operate, and trust.
Agentic Commerce Control PlaneFrom client scope to evidence-backed execution
operational model
Client configurationBrand overlay, users, tools, skills, authority, knowledge.
→
Isolated runtimeScoped package, vault, hostname checks, client-private context.
Slack operating surfaceSetup commands, priorities, status, work requests, receipts.
→
Commerce executionShopify, lifecycle, ads, CRO, reporting, approvals.
20+User-confirmed seven- and eight-figure ecommerce portfolio scope
43Audited agent-factory commits authored
340Audited private Shopify growth commits authored
2Merged client-command PRs with recorded canaries
32 / 65Reusable skill definitions / asset components tracked
The system
Agents need an operating model, not a bigger prompt.
The production work lives around the model: client boundaries, configuration, tools, permissions, team surfaces, evidence, recovery, and the point where a person must decide.
Control plane
Client records, shared kit, brand overlays, launch governance, and operator truth.
Isolated runtime
Scoped client package, private context, secrets, workspace identity, and RLS.
Team surface
Slack-native setup, priorities, requests, progress, approvals, and receipts.
Commerce tools
Shopify, Klaviyo, paid media, analytics, CRO, creative, and reporting.
Evidence loop
Canaries, source states, readback, failures, recovery, and reusable learning.
What I build
One operating layer across growth and infrastructure.
My work sits where agent architecture meets the real ecommerce backlog: paid media, offers, landing pages, Shopify, CRO, lifecycle reporting, client onboarding, and execution.
control plane
Agent infrastructure
Repeatable client launches without collapsing every brand into one shared context.
- Client configuration and brand overlays
- Isolated runtimes and scoped packages
- Secrets boundaries, RLS, identity checks
- Receipts, status, rollout rings, and recovery
ecommerce
Growth execution
Turn recurring ecommerce work into reusable workflows instead of one-off AI tasks.
- Paid-media operations and launch workflows
- Offers, landing pages, advertorials, and CRO
- Modular PDPs, Shopify implementation, and attribution
- Creative briefs, proof systems, and reusable components
truth
KPI reporting and evidence
Reports that know when the data is incomplete and show their work.
- Shopify and Klaviyo KPI output
- Slack delivery receipts and readback
- LIVE, PARTIAL, source, fixture, and target states
- Missing-data reasons instead of silent zeros
adoption
Team-facing agent UX
Lower-friction ways for growth teams to use the system without learning the provider internals.
- Slack-native onboarding and operation
- One-step connector and ads menus
- Reusable skills and brand-local knowledge
- Human-readable priorities, status, and approvals
Evidence-backed work
Proof that the system moved beyond a slide deck.
These examples are intentionally bounded. They show implementation, launch, reporting, and recovery without inventing revenue attribution or claiming every configured agent is live.
Client-safe setup commands
Complex connector and ads setup became a simpler Slack-facing action while preserving readiness gates and client boundaries.
- Two merged PRs
- Client-safe connector and ads aliases
- Recorded live menu canaries
Boundary: current fleet-wide rollout was not revalidated.
KPI reporting with truth states
Shopify and Klaviyo data moved into structured Slack reports with delivery receipts and explicit incomplete-data states.
- Live send and Slack readback
- Repeated daily KPI outputs
- LIVE and PARTIAL status handling
Boundary: reporting operation is not a revenue-lift claim.
Private Shopify growth stack
Direct implementation across modular PDPs, mobile shopping paths, product imagery, proof cards, conversion copy, and attribution surfaces.
- 340 audited authored commits
- 29-definition metaobject setup system
- Versioned advertorial/listicle funnels
Boundary: private implementation is anonymized and not a causal revenue claim.
GrowthOS Skills + Legos
Offers became the hub connecting pages, ads, products, email, and creative learning through a reusable vocabulary.
- 32 skill definitions
- 65 asset-component definitions
- Page identity, structure, offers, and reuse logic
Boundary: shared workbook; not every row is claimed as deployed.
Offer to page to Meta
A live ecommerce funnel moved through offer and page iteration, Shopify and Loop constraints, creative coordination, Meta setup, and launch workflow.
- Public page verified
- Offer and checkout constraints resolved
- Meta setup and launch work completed
Boundary: no ROAS, CAC, CVR, spend, or revenue result was located.
Duplicate app recovery
When setup created duplicate Slack apps, the duplicates were blocked, the correct app identities were restored, gateways restarted, and live canaries recorded.
- Read-before-create lesson
- Immutable identity checks
- Secret-safe recovery path
Senior systems work includes knowing how the automation can fail.
Implementation footprint
Direct work, shared systems, clear boundaries.
The most useful proof separates what I authored from what the larger system contains and what still needs fresh live validation.
Proof
What it establishes
Boundary
43 commitsAgent-factory implementation
Includes 16 commits on client, registry, and skill paths. Does not imply sole factory authorship.
340 commitsPrivate Shopify implementation
Shows direct PDP, creative, attribution, and conversion work. Does not prove causal business lift.
16 configsConfigured brand and in-house systems
Configurations span pilot, internal, and production-candidate states. They are not presented as 16 live agents.
100 tasksCross-functional operating breadth
Audited sample includes paid media, CRO/offers, agent systems, and KPI work. Counts overlap.
32 + 65Reusable skills and components
Shows taxonomy and operating vocabulary. It does not imply every item shipped to every brand.
What I would do inside a brand
A 90-day path from AI experiments to an operating system.
The first goal is not maximum autonomy. It is a trustworthy production loop with clear value, useful team interfaces, and a path to expand.
Map the work and the risk
Audit the recurring backlog, data, tools, handoffs, approvals, and commercial constraints. Prioritize workflows by value, frequency, readiness, and failure cost.
- Brand operating brief
- Workflow and authority map
- Access and data-quality plan
- Top-three deployment roadmap
Ship the first production loops
Deploy a narrow set of useful agents with real tools, observable output, human approvals, and deterministic evidence.
- Client/team operating surface
- First connector and KPI workflows
- Dry runs, canaries, and receipts
- Failure and escalation paths
Connect, teach, and expand
Turn isolated wins into a shared control plane, train each role on the system, measure adoption, and prioritize the next quarter.
- Role-specific SOPs and reviews
- Reusable skills and brand knowledge
- Operator dashboard and health checks
- Quarter-two automation backlog
Team adoption
The system should make the team stronger, not just busier with AI.
Enablement model
Teach people where to trust, review, override, and improve the agents.
My verified history includes adoption-oriented Slack UX and lower-friction operating flows. The next step inside a brand is a deliberate training program tied to real roles and measurable workflows.
Role map
Define how media buyers, creative strategists, lifecycle marketers, ecommerce managers, developers, and leaders interact with the system.
Operating surface
Give the team a simple place to request work, see progress, approve high-risk actions, and inspect the evidence.
Review standards
Document what an acceptable brief, page, report, launch, or recommendation looks like and when a person must intervene.
Learning loop
Capture corrections and outcomes into reusable skills and brand-local knowledge instead of repeating the same feedback forever.
Adoption metrics
Measure usage, cycle time, throughput, review burden, failure rate, and business outcomes without pretending automation itself is the goal.
Commercial ecommerce background
The agent work is built on operating experience.
Before the current agent-factory work, I led ecommerce, performance marketing, and AI automation across brands and client portfolios.
Performance marketing systems
$28M+Meta-attributed revenue directed across ecommerce accounts, with user-confirmed improvements across ROAS, CVR, LTV, email revenue, and CAC.
Director of Ecommerce
$20M → $30MUser-confirmed ecommerce run-rate movement in 90 days, alongside conversion and international growth work.
AI marketing automation
$20K → $120KUser-confirmed MRR growth in 60 days, with automation across creative testing, segmentation, QA, reporting, and funnel optimization.
Production standards
The boring controls are what make the exciting system usable.
Client isolation
One client's data, credentials, context, and outputs must not bleed into another's.
Read before create
Verify identity and existing state before provisioning apps, workspaces, or production objects.
Missing is unknown
Incomplete inputs stay PARTIAL or unknown. They do not silently become zero or complete.
Proof types matter
Source code, runtime output, merge state, rollout state, and business outcomes are different claims.
Human authority
Credentials, spend, billing, publishing, and irreversible changes remain explicitly governed.
Agent systems roles
Bring me the growth team. I will build the operating layer around it.
I am interested in ecommerce leadership roles where I can design, deploy, manage, and expand agentic systems across paid media, creative, Shopify, CRO, lifecycle, reporting, and team operations.
Best-fit environment
A seven- or eight-figure ecommerce brand with real paid-media and lifecycle volume, recurring cross-functional work, leadership support, and a team ready to move from scattered AI usage to a production operating model.