Agentic GTM workshop Part 1: two pages and a reusable skill

On October 8, I built an account page for Zuora with Codex. I saved the checked instructions as a skill and used them for Stripe. That was the hands-on goal of Part 1 of my Agentic GTM workshop.

The project held two local page drafts, source-linked research and instructions the agent could use again. Below is the replay, the five workshop sections, actual screenshots and answers to questions attendees asked.

Watch the full Part 1 recording

The replay includes the live build, troubleshooting, Q&A and the Agency video breaks. Use the chapter links to follow a particular exercise.

What is agentic GTM?

Agentic GTM uses an AI agent to carry out a sequence of go-to-market tasks with company context, tools and checks. In this workshop, the sequence was research, account-page creation, review and reuse of the instructions.

A useful instruction names the output and the evidence needed to accept it. “Build a page for this account” is a start. “Use these company facts, cite the source of each claim, and open the page on desktop and phone” gives you something to inspect.

Our agentic GTM guide explains the broader approach. This session worked through one build, one step at a time.

The five workshop sections

Scroll the table to see each output and check.

1. See working examples

I opened an existing Zuora example that connected an account page, an email and a gift invitation. The page gave different buying roles a reason to explore the offer.

In Part 1, we kept outreach paused while building and checking two unpublished pages.

2. Set up your company

Start with the workshop starter project. Open it in your local coding agent, then use the first prompt in the participant booklet.

Fill in your website, buyer segment, offer and two target domains. Read the generated COMPANY.md before moving on. The agent needs your product, your buyer and a source-supported offer.

For Part 1, use your usual coding-agent access and a local Python environment. The project supports macOS and Linux, with WSL for Windows. Native Windows has not been verified.

Adapted company-review prompt: this is a short version of the workshop instruction. The booklet contains the complete prompts.

Read our company brief. Check the product, buyers, offer and proof against the original sources. Remove unsupported claims. Save the corrected brief and show what changed. Stop after this step.

One useful moment came when the agent asked me to support a CAC reduction claim. I supplied a customer case study. The check still removed the unsupported number from the usable offer.

Evidence for lower cost per click does not establish the same reduction in customer acquisition cost. The lesson is to check what a source supports.

We also checked AUTHORITY.md and AGENTS.md. These files saved the operating rules, including one-step execution and paused sending. PROGRESS.md recorded the completed work.

3. Choose two target companies

I used Zuora and Stripe as the exercise targets. The agent read their official websites, saved a brief for each account and updated TARGETS.md.

For Zuora, the brief connected quote-to-cash positioning with finance and technology buyers. For Stripe, the page later separated enterprise payments from SaaS billing.

Keep facts and proposed campaign ideas distinct. A product description can come from a source. An audience angle is your hypothesis to review.

Each account brief should show the product, likely buyers, source URLs, relevant excerpts and read dates. That gives the next step a basis for its copy.

What is an ABM landing page?

An ABM landing page is a page tailored to a named account or account group. It connects your offer to that account’s business and buyers. In Part 1, Zuora and Stripe were targets for two unpublished workshop drafts.

4. Build and check your first page

The agent used the company brief and Zuora research to write local HTML. The page led with “Reach the people behind quote-to-cash.” It proposed separate campaign ideas for finance and technology leaders.

The page should give that account a reason to consider your offer. Tie the message to the buyers it needs to reach.

Then we checked the copy and opened the page in a browser. The review covered a desktop viewport at 1440 × 900 and a phone viewport at 390 × 844. It checked layout, horizontal overflow and the demo links.

Adapted page-review prompt:

Check every factual claim against its original source. Label proposed campaign ideas. Open the local page on desktop and phone. Test the CTA links. Fix unsupported copy and layout issues, then save the review.

Open the page yourself. Read the offer and test the next step. Our AI marketing workflows guide explains why a usable draft needs sources and an owner.

How do I create a skill in Codex?

A skill saves repeatable instructions in a SKILL.md file. In the workshop, we saved the reviewed page-building process, then asked Codex to read that file and apply it to the second target.

5. Save the skill. Build page two.

After checking the Zuora page, I asked the agent to save the instructions in company-motion/skills/account-page/SKILL.md.

The saved skill told the agent to restore company context, read the target brief, use sourced claims, build locally and check the page. Those checks belonged in the reusable instructions.

Adapted reuse prompt:

Read the saved account-page skill and the Stripe brief. Build and check the Stripe page. Preserve the first page. Keep both pages unpublished. Show the files and review evidence.

The second page used Stripe’s business context. Its headline was “Match the message to the business model,” with separate ideas for enterprise payments and SaaS billing.

The build also hit an ordinary problem. Around 1:19:00, the preview link failed because the local server had stopped. I gave the agent a screenshot of the error and asked it to repair the preview.

After the repair, I opened the page again. Keep the preview server running while you review the page.

Finally, we checked the company brief, target briefs, both HTML pages, saved skill and review files. We kept the project for the next session.

The demonstrated reuse was an explicit instruction to read the saved file. You should also tell the agent to read current project files when resuming work.

Questions attendees asked

These questions and answers are paraphrased from the live Q&A. I answered the questions below. The replay times are approximate.

What tech stack do I need for Part 1?

Codex was the core tool in the live build. Part 1 needs your coding agent’s normal model access, the workshop project and a local Python environment. It does not require additional enrichment, email, gifting or advertising API keys. Watch at 42:21.

Do I need my own GitHub repository?

You can use your own repository or keep the project locally. I said both approaches were possible. The public starter project also has a ZIP download, so you can open the unzipped folder in your coding agent. Watch at 47:00.

Is Humanizer built into this project?

Humanizer is a separate skill for editing AI-sounding prose. During the Q&A, I said I would check whether it was bundled. The starter ZIP linked here currently does not include it. Watch at 46:40.

Can I use Perplexity Computer instead?

I said this should be possible, but I did not demonstrate it. The recorded build uses Codex. Check whether an alternative can read the project, write local files and preview the pages before relying on it. Watch at 42:06.

Can these pages live in HubSpot?

Yes was my answer in the session, and I suggested HubSpot MCP as an integration route. Hosting is a separate step from generating local HTML. We did not deploy these workshop pages into HubSpot. Watch at 48:36.

Can I customize the landing pages?

Yes. Ask the agent to change the content or layout, then repeat the source and browser checks. I also suggested requesting an inline editor. We did not build or demonstrate that editor in Part 1. Watch at 1:02:18.

Can the agent find lookalike accounts?

Give it explicit criteria, such as company size, buyer profile and product category, plus sources to research. I suggested starting with customer websites and G2. We discussed the approach without generating a lookalike account list in this session. Watch at 1:11:15.

How would I add an interactive ROI calculator?

I suggested an HTML calculator and asking the agent which inputs would matter to your prospects. We did not build one during the workshop. Any calculator would need checked inputs and assumptions before you use its estimates externally. Watch at 1:32:10.

Where can I get the recording, slides and project?

The recording, slides and participant booklet are on the workshop site. Use the password from your workshop email. The starter project is also available as a public GitHub ZIP. Direct links appear below. Watch at 42:13.

Keep building with Parts 2 and 3

In the closing preview, Part 2 moved toward a small email pilot: choose accounts, check buyers, preview exact messages and send a test to yourself. Part 3 added gifts, LinkedIn and ads to the workflow.

Bring the project you built in Part 1. Start with two pages you have opened and checked before expanding the account list or connecting delivery tools.

For connected advertising workflows, see the Metadata MCP recipes and platform. Use those recipes when you are ready to connect advertising accounts.

Workshop materials

• Download the starter project.
• Open the participant booklet.
• Open the Part 1 slides.
• Watch the complete live recording.

Source and review scope

This article covers my October 8, 2026 workshop only. Gemini 3.8 Flash processed its complete audio and moving video, including demos, video breaks and Q&A. Codex verified coverage, inspected the selected screenshots and checked relevant passages against the produced captions. The account pages are unpublished workshop drafts. The session does not establish attendee outcomes or campaign results.