GTM agents · the playbook of an outbound agency
GTM agents: how frontBrick builds outbound with BlitzAPI
The short answer
GTM agents are AI agents that each own one stage of go-to-market work, under an orchestrator that holds the brief, the spend and the launch. At frontBrick, four GTM agents run outbound: Strategy sets the ICP and sizes the market, List builder pulls and qualifies it with BlitzAPI, Copywriter writes the sequences, and Upload loads them for an explicit go.
On this page16 sections
- What is a GTM agent?
- How do GTM agents work together?
- How do you size a market before writing the brief?
- How do you build an outbound list with AI agents?
- Try it: what does checking fit first save?
- How do AI agents write outbound sequences without inventing leads?
- When should an AI agent launch a campaign?
- How do you test GTM agents before you trust them?
- What does a GTM agent need from its data layer?
- Who is frontBrick?
- How do you call TAM by People?
- How do you find the decision maker in each company?
- Which signals are filters, and which come back as data?
- How do agents stay under the rate limit?
- How do you connect an agent to BlitzAPI?
- Sources and method
frontBrick is an outbound and GTM agency and a BlitzAPI customer. BlitzAPI is the data layer under almost every list it builds. This guide walks through the playbook frontBrick actually runs: one orchestrator, four specialist agents, and real leads from the first row to the sequencer.
frontBrick also published its take on BlitzAPI for GTM teams. What follows is the live agent flow behind that page.
What is a GTM agent?
A GTM agent is an AI agent that owns one job in your go-to-market: sizing a market, building a list, writing a sequence. It takes a brief, does that job with its own tools, and hands a defined output to the next agent.
- Definition GTM agent also called go-to-market agent
- An AI agent that owns one job in a go-to-market motion, such as sizing a market, building a list or writing a sequence, and hands its work to the next agent.
PreciselyA GTM agent is an AI agent scoped to one stage of go-to-market work. It takes a brief, calls the tools and data sources of that stage, and returns a defined output for the next stage. An orchestrator keeps what no single agent should own: the shared brief, the spend and the launch.
Keeping agents specialized gives every stage an output you can check. A list is clean or it isn’t. A sequence uses real leads or it doesn’t. An AI SDR usually sells that whole chain as one product; GTM agents are the pieces you assemble yourself, so you choose the data, the rules and the launch.
When a whole motion runs this way, it is called agentic GTM, or AI GTM. A GTM engineer usually designs the agents, the rules between them and the data they can reach.
How do GTM agents work together?
Under one orchestrator, as one outbound system. At frontBrick, the Orchestrator holds the brief, the spend and the launch, and four agents do the stages. BlitzAPI is what they reach for first whenever companies, people or enrichment are needed.
- Definition Agent orchestrator also called orchestrator agent
- The agent in charge of the other agents: it holds the brief, decides who works next, and keeps control of the spend and the launch.
PreciselyAn agent orchestrator coordinates specialist agents in a multi-agent system. It assigns each stage, passes outputs from one agent to the next, and holds the approvals no specialist can grant itself, such as budget and launch.
- Agent 1StrategySets the ICP, the offer, the angle, the channels and the TAM.Hands overCampaign brief
- Agent 2List builderPulls the market, checks fit, fills emails and extra data.Hands overValid, clean list
- Agent 3CopywriterWrites the sequences from the brief and the clean list.Hands overFinished sequences
- Agent 4UploadLoads them into the sequencer, for email and LinkedIn.Hands overCampaign, waiting for the go
How do you size a market before writing the brief?
You look at the companies that actually employ your buyer, before the brief is final. frontBrick’s Strategy agent owns the thinking before anyone pulls a row:
- Who they sell to, who they exclude, and why this angle lands now.
- Which channels they run: email, LinkedIn, or both.
- A TAM by People read from List builder, so volume is grounded in accounts that actually employ the buyer persona.
frontBrick does not invent market size. It looks at who employs the buyer.
TAM by People starts from the person, not the company. It finds the people who match your ideal customer profile, then returns the companies they work for, each one once. Strategy uses that number to keep the brief honest, and the brief locks only when the market size is clear. Companies and people stay two separate counts, so nobody mistakes contacts for accounts.
People who match the persona
- Head of Customer SuccessNorthwind
- VP of Customer SuccessNorthwind
- Head of Customer SuccessHarbor Labs
- Head of Customer SuccessAlder Systems
Companies, each once
- Northwindmatched_people: 2
- Harbor Labsmatched_people: 1
- Alder Systemsmatched_people: 1
When the better signal is what a company is hiring for, rather than who already works there, the same idea runs on job postings. The TAM guide walks through it.
How do you build an outbound list with AI agents?
Go company first: broad, then narrow, and look for people only after fit. This is the stage most people mean by AI agents for lead generation. It is where most of the work happens at frontBrick, and where BlitzAPI earns its keep. List builder uses TAM by People to pull the companies where ICP people work: thousands of them in minutes, with descriptions, firmographics and domains in the same response.
- Pull the marketTAM by People, broad on purposeBlitzAPI
Thousands of companies
- Check ICP fitReads each company descriptionJev by TypeSafe AI
Fit companies only
- Find decision makers and emailsWaterfall ICP, only on the companies that fitBlitzAPI
The people kept
- Fill the gapsMissing emails, competitors, tech stackClay
A valid, clean listNo row skipped the fit check or the email check
Each tool has one job in that flow, with BlitzAPI at the center:
| Job | Who does it | How frontBrick uses it |
|---|---|---|
| Pull and enrich companies | BlitzAPI | Descriptions, firmographics and identity land in one pass. |
| Check ICP fit | Jev by TypeSafe AI | Reads the BlitzAPI company descriptions and judges each account against the ICP in minutes. Fit companies move on. |
| Find decision makers | BlitzAPI | Waterfall ICP searches each fit company for the titles in order and keeps the best available person. |
| Find emails | BlitzAPI | Email Enrichment already covers a lot of valid emails on the people kept. |
| Fill email gaps | Clay’s CLI | A waterfall adds more emails and validates them. |
| Extra enrichment | Custom Clay functions | Data from BlitzAPI feeds functions that find competitors and check tech stack. |
Two different waterfalls sit in that flow. Waterfall ICP picks the right person in a company, as the guide to finding decision makers explains. Clay’s waterfall looks for that person’s email.
The output is a valid, clean list. No row reaches the Copywriter if it skipped the fit check or the email check.
A big plus for frontBrick: ad spend, funding, open jobs and new hires come straight from BlitzAPI, so it doesn’t need Clay for them. These buying signals decide which segment List builder pulls. Companies that raised recently, or that spend on ads, are a filter on the search, so every row in that segment already carries its reason to reach out.
Try it: what does checking fit first save?
The order of the steps decides what a list uses. Every company, person and email the API returns counts as one record, so checking fit before you look for people spends records only on accounts you would actually write to. Move the sliders to your own numbers.
Try it · what does checking fit first save?
Enrich every company25,000 records
- 5,000 companies
- 10,000 people
- 10,000 emails
Check fit first11,000 records
- 5,000 companies
- 3,000 people
- 3,000 emails
Checking fit first uses up to 14,000 fewer records on this list. Start covers 18 lists like this a month, Scale covers 454 lists like this a month.
The figures are upper limits in BlitzAPI records: a lookup that finds nothing counts nothing. The defaults are examples, not measurements. Your fit check, your email waterfall and your sequencer bill on their own.
How do AI agents write outbound sequences without inventing leads?
They write only from real rows. The valid, clean list goes to the Copywriter agent with the campaign brief from Strategy, and those two inputs are what it works from.
- Real leads only. The Copywriter does not invent people to fill a sequence.
- Context from the row. Company and person context on each row started in BlitzAPI.
- A reason to reach out. Signals like funding, open jobs and new hires are already there when copy needs one.
- A preview first. frontBrick previews a few real leads before scaling the full list.
One row of the clean list
- AngleFrom the campaign briefStrategy
- CompanyNorthwind, Software Development, 51-200 peopleBlitzAPI
- ICP fitFit company, moved onJev by TypeSafe AI
- PersonMaya Chen, Head of Customer SuccessBlitzAPI
- Work emailmaya@northwind.exampleVerifiedBlitzAPI, Clay for gaps
- SignalPulled in the recent funding segmentBlitzAPI filter
- ExtrasCompetitors, tech stackCustom Clay function
When should an AI agent launch a campaign?
Only after an explicit go. When the sequences are ready, they are uploaded to the sequencer for email and LinkedIn, with the leads that survived BlitzAPI, Jev and the email checks. Then everything waits.
Strategy approval is not send approval.
Strategy approval and list spend are not the same as launch. frontBrick still waits for an explicit go before anything enrolls, and the Orchestrator holds that decision, not the agents doing the work.
- Strategy approvedDoes not send
- List spend approvedDoes not send
- Preview on a few real leadsDoes not send
- Explicit goThe only step that enrolls anyone
How do you test GTM agents before you trust them?
On a mixed sample, before real volume. Run the flow on known customers, plausible prospects, wrong-fit companies and ambiguous names, then inspect what comes back:
- Company identity. Is this the right company?
- Current employer. Does this person still work there?
- Persona relevance. Is this the role the brief asked for?
- Contact status. What came back for this person, and is it verified?
A high match rate is not useful if the matches are the wrong people. So frontBrick compares final accepted records, not database sizes, and keeps the denominators honest: email coverage among accepted people is a different number from email coverage across every raw search result.
What does a GTM agent need from its data layer?
Four things: coverage, enrichment in one place, speed and reach. Agents only work with what you feed them, so these decide how good every stage above can be. They are also why BlitzAPI is the backbone of frontBrick’s outbound stack, with thousands of cold emails riding on it every single day.
- Coverage. People and company data that holds up in real list builds.
- Enrichment in one place. The enrichment endpoints sit together, including the signals frontBrick used to chase elsewhere.
- Speed. Agentic pulls come in bursts, so request rate is crucial. frontBrick runs on Scale: $499 a month, 25 requests per second per endpoint and 5,000,000 records a month. Each endpoint has its own budget, so email lookups don’t slow down people searches.
- Reach. The same layer works across Grok Bot, Hermes, Claude Code and Codex, through the API or the BlitzAPI MCP server.
Running the same system for several clients? BlitzAPI for agencies covers how that scales, and the guide to AI agents and B2B data covers how agents reach the data.
Who is frontBrick?
frontBrick is a cold email lead gen agency for outbound and GTM, led by Stan Stojanovic. It builds agentic list systems around BlitzAPI, not one-off CSVs.
For builders
The playbook above is frontBrick's. This part comes from the BlitzAPI team, for whoever wires the agents: the calls behind each stage, run on November 7, 2026.
How do you call TAM by People?
One request takes people filters and company filters together, and returns companies:
curl -X POST https://api.blitz-api.ai/v2/company/tam-by-people \
-H "x-api-key: YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"people": {
"job_title": {"include": ["Head of Customer Success"]},
"location": {"country_code": ["US"]}
},
"company": {
"employee_range": ["51-200"],
"keywords": {"include": ["software"]}
},
"max_results": 50
}'A result looks like this, with a fictional company in place of the real one and the long text fields cut:
{
"results": [
{
"company": {
"linkedin_url": "https://www.linkedin.com/company/northwind",
"name": "Northwind",
"about": "Northwind builds software for...",
"specialties": ["saas", "automation"],
"industry": "Software Development",
"type": "Privately Held",
"size": "51-200",
"employees_on_linkedin": 140,
"founded_year": 2017,
"hq": { "city": "Boston", "country_code": "US" },
"domain": "northwind.example"
},
"matched_people": 1
}
],
"results_length": 1,
"max_results": 50,
"cursor": "eyJhIjp7..."
}Each result is one company and one record, and matched_people says how many of its people match. Send cursor back for the next page. There is no total field on this endpoint: to get the size of the market, page until cursor is null. Pagination stops at 50,000 companies, so split a wider market into narrower filters. When firmographics alone define the market, Company Search returns total_results on the first page.
How do you find the decision maker in each company?
Company first: take the linkedin_url of each fit company and send it to Waterfall ICP with the titles in order. It tries the first level and moves down only while slots are left.
curl -X POST https://api.blitz-api.ai/v2/search/waterfall-icp-keyword \
-H "x-api-key: YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"company_linkedin_url": "https://www.linkedin.com/company/northwind",
"cascade": [
{"include_title": ["Head of Customer Success"]},
{"include_title": ["VP Customer Success", "Director of Customer Success"]},
{"include_title": ["CEO"]}
],
"max_results": 1
}'Each result carries icp, the level that found the person, ranking, and the person with the job that matched. On this run, the first level answered. Waterfall ICP returns no email: send person.linkedin_url to Find Work Email, POST /v2/enrichment/email, which counts a record only when it finds a verified address.
Which signals are filters, and which come back as data?
A filter narrows the pull; it doesn’t add a column. Ad spend and funding values aren’t in the response, so a company is in the segment because it matched.
- Ad spend: a filter,
company.ad_spend, on estimated monthly Google Ads spend. You get the companies that match. - Funding: filters,
company.last_funding_type,last_funding_year,last_funding_amountandtotal_funding. You get the companies that match. - New hires: a filter,
people.is_recent_hire, for a job that started in the last 90 days. Waterfall ICP takes it in eachcascadelevel, and returnsjob_start_dateon the matched job. - Open jobs: data. Jobs Search, or
POST /v2/jobs/companyfor one company, returns the postings themselves.
To give the Copywriter its reason to write, run one pull per segment and tag the rows yourself. Many companies have no public ad spend signal, so that filter narrows fast: pair it with a broad industry.
How do agents stay under the rate limit?
The limit applies per endpoint and per account: all your API keys share each endpoint’s budget. Read yours in max_requests_per_seconds from GET /v2/account/key-info, and set rate_limit_rps in the SDKs to match, since their default of 5 leaves room unused on a paid plan. A call above the limit returns 429 at once, with Retry-After: 1.
How do you connect an agent to BlitzAPI?
Through the MCP server: one URL, one sign-in, one tool per endpoint. In Claude Code:
claude mcp add --transport http blitz-api https://api.blitz-api.ai/mcp
claude mcp login blitz-apiCodex and Grok take the same URL, and the documentation has each command. MCP suits lookups in a chat and writing the scripts; for a pull of thousands of companies, have the agent write a script against the REST API.
Sources and method
- Written by
- The BlitzAPI team
- Playbook by
- Stan Stojanovic, Co-Founder & CEO, frontBrick
- Reviewed by
- Antoine Blitz, Founder, BlitzAPI
- Published
- API calls run on
- BlitzAPI for GTM teams: pricing, workflows and implementation, frontBrick
- TAM By People, BlitzAPI API reference
- Waterfall ICP Search, BlitzAPI API reference
- Company Search, BlitzAPI documentation
- Plans, records and rate limits, BlitzAPI documentation
- MCP server, BlitzAPI documentation
- Clay API and CLI, Clay documentation
This guide documents frontBrick's playbook, as shared by Stan Stojanovic. frontBrick is a BlitzAPI customer. The BlitzAPI team wrote this version, checked the product facts and added the simulator and the For builders section.
Jev is a product of TypeSafe AI, and Clay is a separate platform. Both are part of frontBrick's own stack, not BlitzAPI features, and each bills on its own.
The BlitzAPI team ran the API calls on November 7, 2026, on real data. In the sample calls, a fictional company stands in for the real one; the field names and the structure are kept as returned.
Northwind, its people and the other companies in the diagrams are fictional. The simulator's defaults are examples, not frontBrick measurements.