How to connect PostHog to an AI agent: what GTM AI agents can do with PostHog in Day AI
Updated
The short answer. To connect PostHog to an AI agent, sign each person in to PostHog's MCP server with their own login and choose which PostHog tools may run without asking, so GTM AI agents, which put AI for GTM to work account by account, can check how a customer uses the product and act on flags and surveys from the same place they read the customer's calls and email. In Day AI, anyone in the workspace connects PostHog from Settings > Connectors with their own login, no Owner setup required, and their agent uses PostHog's MCP server to query product analytics, investigate errors, and manage feature flags, experiments, dashboards and surveys, whether you use Day AI in place of your CRM or on top of Salesforce or HubSpot.

Let's say you run customer success at a developer-tools company, and a call with your biggest account just ended at 3:15. Their lead engineer asked twice whether the new bulk export is ready, and mentioned that something in the dashboard has been erroring since Monday. You open Day AI and ask your agent what you can do today. It reads the call, then checks PostHog: bulk export sits behind a feature flag that isn't on for this account, and there is a cluster of errors on the dashboard page from their users this week. The agent proposes turning the flag on for the account, which appears as an approval card in the chat, and drafts a reply that confirms export is live and that the error is with the product team. You approve the flag, tweak one line of the email, and send it before their engineer has closed the meeting notes.
Where the usual answer goes wrong. PostHog AI setups tend to answer product questions for the product team: what's trending, which funnel step leaks. The customer-facing team needs the same data turned around, one account at a time, next to what that account said. Day AI starts from the conversations and treats PostHog as the place the agent checks and, with a person's approval, changes. If your success team already pings an engineer to check a flag or an error for a customer, that habit was right. Now the person on the call can check it directly.
The flag, feedback, follow-up loop for PostHog, with Day AI
PostHog becomes useful to a go-to-market team when product facts reach the person talking to the customer. The loop has three parts.
- Flag: what the account can see. Whether a customer has a feature depends on flags and experiments in PostHog. In Day AI, your agent looks up the flag or experiment for an account in chat and, once you approve, can change it.
- Feedback: what the account is running into. Usage drops, errors and survey answers explain a lot of what customers say on calls. In Day AI, the agent reads PostHog's analytics, errors and surveys next to the account's meetings and email, so you see both at once.
- Follow-up: what you tell the customer. A fix nobody mentions is a fix the customer never learns about. In Day AI, the agent drafts the note telling the customer the feature is on, right after you approve the flag.
How to connect PostHog to Day AI
Step 1: Connect your PostHog account in Day AI
Go to Settings > Connectors, find PostHog through search or Browse connectors, and click Connect. Sign in to PostHog and approve access. No workspace Owner has to set anything up, so anyone in the workspace can connect their own account, and each person's agent sees only what their PostHog login can see. Day AI then loads the tools PostHog's MCP server offers.
Step 2: Choose what Day AI may do in PostHog on its own
Open the PostHog connector and review Tool permissions. Set all on the Write & delete group is the quick way to keep every PostHog change at Needs approval. Lookups such as analytics queries and error searches start at Always allow. Anything that changes PostHog, like editing a flag, an experiment or a survey, starts at Needs approval. If there's a tool you never want an agent to use, deletes for example, set it to Off.
Step 3: Ask your Day AI agent about one account
Try questions that mix product and conversation. "Is the new export flag on for Acme, and did they ask about export on any call this month?" "What errors have Acme's users hit this week?" "How did Acme answer our last PostHog survey, and does it match what they said in Thursday's QBR?" Your agent checks PostHog and reads the account's history in Day AI before it answers.
Step 4: Approve PostHog changes in the Day AI chat
When the agent proposes a change, an approval card appears in the chat naming PostHog and the action, with Approve and Deny. The change happens only after you approve. The same goes for the customer email the agent drafts: it lands in your drafts for you to read and send.
Step 5: Give a Day AI skill the PostHog reads
For a weekly view, ask your agent to create a skill, such as a summary before each QBR of errors and usage changes for that account, and select PostHog in the skill's Connectors picker. Nobody is there to approve a flag change while a skill runs, so give the skill PostHog's reads and make flag changes in chat. You can see every Day AI integration and connector to pair PostHog with the rest of your stack.
Where routing product questions through engineering breaks
- The answer arrives after the call. A Slack message to an engineer gets answered tomorrow. In Day AI, the person on the call asks the agent and gets PostHog's answer in the same chat.
- Context gets lost in the handoff. The engineer sees a flag name, not the customer who has asked about it three times. In Day AI, the agent reads the account's calls and email next to the PostHog data.
- Changes have no trail to the customer. A flag flipped for a customer often goes unmentioned to them. In Day AI, the agent drafts the follow-up in the same chat where you approved the change.
- Everyone needs a dashboard. Building a view per account doesn't scale past a handful. In Day AI, a skill reads PostHog for every account on your list and reports what changed.