AI agent developer for marketing: a 19-agent stack, with approval before anything spends or publishes.
I build AI marketing agents and bots on a 19-agent Claude and MCP stack, with a person approving anything that spends or publishes. I run that stack with three people for reporting, audits and research. Agentic workflows, custom GPTs and n8n-style automation are scoped on the automation page when that is the job.
TL;DR
An AI agent developer builds a system that finishes a multi step job: pull the numbers, check the rules, draft the output, and wait for a person.
I run 19 of those agents with three people, on Claude and MCP, for reporting, audits and research. That is the production stack, not a slide.
Customer facing bots are on the AI chatbot page. Agentic workflows, custom GPTs and n8n-style automation are on the AI automation page. Claude setup is on the Claude consultant page, and the longer note is Claude for marketing teams.
Vicky Lalwani is an AI agent developer for marketing teams. His own stack is 19 agents and 3 people on Claude and MCP, connected to Search Console, GA4, Google Tag Manager and ad data. Customer facing bots are built on the AI chatbot page. Agentic workflows, custom GPTs and n8n-style automation are on the AI automation page. A person approves anything that spends or publishes.
At a glance
Senior judgement, machine speed.
Agents handle the work that eats your week: pulling data, checking it against rules, writing the first draft and flagging what needs a human. Your team spends its time on strategy and creative instead.
Every agent I build has a clear job, a defined data source, written instructions and a human checkpoint. No black boxes.
Bot building is part of the same practice, with a different job. A chatbot answers, qualifies and books on your site or on WhatsApp. An automation moves work between Claude, ChatGPT, MCP and the tools you already use, including n8n-style flows and custom GPTs. Anything that spends or publishes still waits.
- Reporting agents: weekly and monthly reports, written for a person to send
- SEO agents: audits, briefs, QC and internal linking
- PPC agents: search term reviews and anomaly alerts
- Research agents: keywords, competitors and AI answer checks
- Chatbots and WhatsApp bots, built on the chatbot page
- Agentic workflows, custom GPTs and n8n-style automation on the automation page
Who should build the agents?
An agent that sends mail or changes an ad account without a checkpoint is a liability. This is how the options differ.
| This consultancy | An automation agency | A team using ChatGPT ad hoc | |
|---|---|---|---|
| Who builds it | The person who runs the same stack daily | A delivery team you may not meet | Whoever has a spare afternoon |
| Human approval | Required before anything reaches a customer or an ad account | Depends on the statement of work | Usually none |
| Data | MCP connections to the tools you already use | Often a new platform to rent | Copy-paste |
| Handover | Documented so your team can run it | Retainer to keep the lights on | Lives in one person's chat history |
| Bots and workflows | Chatbots and automation scoped as their own builds | Often folded into a larger programme | A prompt saved in one person's chat |
| Best for | Lean in-house teams and agencies in AU, UK or US | Large programmes with many stakeholders | One-off drafts |
Agents built around your workflow
Workflow mapping
Identify the tasks worth automating, with hours saved per month.
Agent design
Clear role, inputs, rules and human approval points for each agent.
Data connections
MCP servers and APIs for Search Console, GA4, Ads, Semrush, CRM and Sheets.
Build & test
Agents tested on your real data until outputs are reliable.
Dashboards
Looker Studio and alerting so you see what agents did and found.
Handover
Documentation and training so your team can run and extend them.
How we'll work together
Map
Audit the week and pick the highest-value tasks.
Build
Design and build the first agents on your data.
Prove
Run in parallel with your team until outputs match.
Scale
Add agents and connect them into pipelines.
Clear terms, no surprises
What working together looks like before you sign anything.
You own everything
Ad accounts, analytics, data, content and assets stay in your name, always, including if we stop working together.
No minimum budget
No minimum ad spend. I'll tell you honestly what your budget can achieve on the free call.
3-month minimum, then monthly
Three months gives the work time to show results. After that it's month to month.
Senior work, direct access
I lead the strategy and I am your contact. A small team handles delivery. My AI agents cover the repetitive audits and reports.
Scoped and quoted up front
Pricing depends on channels and markets. You get a written scope and quote after the free 30-minute call.
Plain-English reporting
Regular reports tied to leads, bookings and revenue, not vanity metrics.
Results from similar work
Client names and domains withheld for confidentiality.
A 19-agent AI marketing team
Claude + MCP agents connected to Search Console, GTM, Semrush and ad data, automating reporting, audits and research.
A dental practice's AI turnaround
Rebuilt the patient-acquisition funnel with paid search, AI booking and lifecycle automation.
Questions people ask
What does an AI agent developer do?
An AI agent developer designs and ships systems that complete a multi step marketing job on your data. I build them on Claude and MCP, with a written job, a data source and a person who approves the output. The 19-agent stack I run with three people is the working example.
Are you an AI automation consultant?
Yes. On this page the job is the marketing agent: audits, reports, search terms, research. The wider build, including custom GPTs and n8n-style flows that move work between tools, is scoped on AI automation. A person still approves anything that spends or sends.
What is an agentic AI consultant?
Someone who wires a workflow so it can take the next step, and who writes the stop rule. I do that for marketing teams. The agent fetches, checks and drafts. It does not email a customer, change a bid, or publish a page until a person says so.
Can you build AI agents for a marketing team?
Yes. A typical first agent is a weekly report or a search term review. A full stack is built in phases. Your team keeps the accounts. I document the job so you can run it after handover.
What are MCP servers for marketing tools?
MCP (Model Context Protocol) is how Claude reads the tools you already use: Search Console, GA4, Google Tag Manager, ad accounts, Semrush. I connect them with the least access that does the job, read only unless you have approved a change. Setup detail is on the Claude consultant page.
Can I hire you as an AI automation freelancer?
Yes. You work with me, for teams in Australia, the UK, the US and the UAE. There is no rate card on this site. The scope is quoted after the free 30 minute call. Customer facing bots are a separate scope on AI chatbots.
What is agentic AI?
Agentic AI is a system that can take the next step itself: fetch the report, apply the rule, draft the action, and stop where you told it to stop. It is not a longer chat reply. If you want that pattern for ops, custom GPTs or n8n-style flows, start at AI automation.
How is an AI agent different from a chatbot?
A chatbot talks to a customer on your site or on WhatsApp: it answers, qualifies and books, then hands over. An agent works inside your team's tools: audits, search terms, reports. Both need a person for anything sensitive. Chatbots are scoped on the AI chatbot page.
What can AI marketing agents automate?
Reporting, SEO audits, keyword and competitor research, content briefs and QC, PPC search term reviews, anomaly alerts and outreach drafts. Strategy, budgets and final approvals stay with people.
Which AI models do you use?
Mostly Claude (including Claude Code and MCP) for analysis and long-form work, and ChatGPT where it fits a team's existing tools. I choose per task.
How long does it take to build an agent?
A focused agent such as a weekly PPC report usually takes one to two weeks to build and validate. A full multi-agent stack is built in phases over a few months.
Will AI agents replace my marketing team?
No. They remove repetitive work so a small team can deliver far more. My own stack is 19 agents and three people.
Do you build AI agents for agencies?
Yes. The same agent can run across every client account, which is where a small agency gets its hours back. A person still approves anything that reaches a client or changes an ad account.
Do you build AI marketing agents for Australian teams?
Yes. The work is remote, with afternoon overlap for AEST and AEDT. Perth calls are booked on AWST. See Australia.
Who approves what an agent sends?
A person does. Agents draft, flag and file. They do not email customers or change bids unless someone has approved that step.
Works best together with
AI Chatbots
Website and WhatsApp chatbots that answer, qualify and book, with a person on the handover.
Learn more →AI Automation
Agentic workflows, custom GPTs and n8n-style automation on Claude and MCP, with a person before anything spends or sends.
Learn more →Claude Consultant
Claude, Claude Code, MCP servers and Skills set up for marketing teams, by a certified Claude practitioner.
Learn more →Analytics & Tracking
GA4, Google Tag Manager, enhanced conversions, Meta CAPI and Looker Studio: measurement you can trust.
Learn more →Let's build something that grows.
Tell me where you want to grow: search, ads, social or AI. I'll reply within one business day with honest next steps.