A meeting comes up where I need to show the work. So I open Notion, update the brain. Open MailerLite, check the campaign metrics. Open the Google Sheet, log the latest pipeline numbers. Open HubSpot, double-check that contact records still match. By the time the meeting starts, every system is current. It feels good for about a week.
Then the systems drift. Notion gets stale. The Sheet falls behind. The email tool has results I never logged anywhere. Another meeting comes up. The cycle restarts.
For years that was the rhythm. I dealt with it because every operator deals with it. The pain wasn't any single tool. It was the recurring update tax I paid every time a stakeholder needed visibility.
What broke the pattern wasn't a better tool. It was AI workflows. The criteria for choosing software changed. The question stopped being "does this tool do its job well" and started being "can Claude actually talk to this directly."
This is the May 2026 cut of the stack that emerged.
The new criteria
Most marketing stacks are evaluated on features. The new criteria is different. A tool earns its place if Claude can read it, write to it, and orchestrate around it without me opening a tab. Each context switch is a tax. A stack designed around one orchestration layer pays it once.
I'm not fully off the old systems. Notion is still where some knowledge lives. Google Sheets still tracks what hasn't been migrated. This is reevaluation in progress, not a clean rebuild. The deeper frame on what this looks like as a system lives in the AI content systems pillar guide.
The stack below is six tools, doing the work that twelve used to do, for sub-$250 a month.
PostHog: analytics one prompt away
PostHog replaced Google Analytics. The reason isn't analytics quality. It's that I connected PostHog to Claude through MCP, which means I can live-prompt for production data during a session. "Show me organic landing pages from the last 28 days." "What's the conversion rate on /contact, broken down by source?" The data is one prompt away, not one tab away.
I set up dashboards the same way. No drag-and-drop dashboard builder. I describe what I want, Claude builds the insight, the dashboard updates. When I want to revise the cut, I prompt the change instead of clicking through the UI.
The shift isn't from a worse tool to a better tool. It's from interrupting myself to look something up to never leaving the conversation in the first place.
Linear: the kanban that survives a Claude session
Sessions in Claude get messy. A two-hour session might generate twenty observations, ten followups, three things I should fix later. By the end the to-do state lives in scrollback, which is the same as not existing.
Linear holds the work that needs to outlive the session. It's where ongoing platform issues live for the Content Lab app, where SEO experiments are tracked for duo.ca, where I separate "today's focus" from "noise to revisit." Claude can open issues from inside the repo when it spots tech debt, which means the cleanup work doesn't get lost in the shuffle.
I just adopted this. Six months from now I'll have a more seasoned take. For now: it's the first project tracker I've used that didn't feel like it needed a separate ritual to maintain.
GitHub, Vercel, and Supabase: the storage layer
Group these three. They're separate tools, but in this stack they do one job. Save anywhere, access anywhere, durably.
GitHub holds the duo.ca codebase, the Content Lab platform code, every MDX article on this site. When I draft an article, it's a file in a repo, version-controlled, reviewable, durable. Not a Google Doc that lives in someone's account.
Vercel deploys all of it. Push to main, the site rebuilds. I can ship a new landing page, a content piece, an experimental redirect, in under a minute. The "rapid landing page experiment" most marketers talk about is real here, because the deploy step is invisible.
Supabase holds the operational data. The Content Lab platform stores client information there. My LinkedIn post archive, my voice profile, my business rules, the structured content that powers the AI workflows live there too. The shift was treating all operational data as something that belongs in a durable store, not a Notion doc that drifts.
The collective mental model is simple. Code is in GitHub. Sites are on Vercel. Data is in Supabase. Claude can read and write to all three. Nothing critical is trapped behind a vendor's UI.
Resend: email infrastructure
Resend handles transactional email. Newsletter delivery, contact form notifications, anything that needs to go from a system to a person.
It replaced parts of MailerLite and HubSpot. Fewer logins, cleaner API, costs almost nothing at the volumes I send. So far a really great solution. I'll have a more developed take after another six months.
What this costs
The full stack runs sub-$250 a month. That includes Claude Code, Google Workspace, Descript for video, and the tools above. Migrating duo.ca off Webflow knocked $40 a month off the recurring bill alone. The savings weren't the goal, but they were a useful confirmation that the new stack wasn't more expensive than the old one.
Most pricing pieces in this category are vague. Mine is exact. Specific numbers are easier to argue with, and easier to trust.
What this means
This isn't a manifesto. HubSpot still serves a real purpose for teams that need a CRM with sales integrations baked in. MailerLite is still the right call if your needs map cleanly to its product. The traditional marketing stack isn't dead.
The point is narrower. AI workflows change the criteria for choosing tools. If your stack can't talk to Claude, you're paying a context-switching tax every time you sit down to work. If it can, the orchestration layer becomes the stack.
For a senior B2B operator running a lean function, that's the difference that mattered. Reevaluating is enough. You don't need to throw it all out.
Common questions.
What is the AI martech stack for B2B operators in 2026?
A lean stack of developer-first tools that Claude can talk to natively. PostHog for analytics, Linear for project tracking, GitHub and Vercel for code and deployment, Supabase for data storage, and Resend for email. The orchestration layer is Claude itself, which removes the context switching cost that defines traditional marketing stacks.
How does Claude connect to tools like PostHog and Linear?
Through MCP servers. Each tool exposes an API endpoint, Claude Code authenticates once, and from then on I can prompt against the data live during any session. PostHog returns dashboards and queries, Linear opens or updates issues, GitHub commits articles. The connection is configured once, not per-session.
Is the AI operator stack worth the cost for a small B2B team?
The full stack runs under $250 a month for me. That replaces a traditional setup that included Google Analytics, Mailchimp or HubSpot, Notion, and Google Sheets, plus the time tax of keeping all of them current. For a small team running a lean B2B function, the math is straightforward.
What replaces traditional marketing tools like Mailchimp or Google Analytics in this stack?
PostHog replaces Google Analytics, with the addition that Claude can query the data directly. Resend replaces transactional email use cases that previously went through Mailchimp or MailerLite. For CRM-style work, this stack assumes you do not need a heavy CRM, which is true for solo operators and small teams but not for sales-led organizations.
Should I move my entire marketing stack to AI-orchestrated tools?
Probably not all at once. I am still on Notion for some knowledge work and Google Sheets for tracking that has not been migrated. The reevaluation happens tool by tool, on the criteria of whether Claude can interact with it. The traditional marketing stack is not dead. The criteria for choosing tools has shifted.
