Scott Martinis

Hey @posthog - I'm trying to find a way to replace @HubSpot on my website for website tracking, (because the Hubspot marketing MCP is expensive and weak)

But there's one thing they do that's a pain to do in posthog.

Deanonymize past web visitors on form fills, ideally calendar form fills.

@Calendly only ships Ga4 and I think maybe Hubspot, and @LumaHQ doesn't allow a posthog pixel.

There are a few forms but nothing standout

How would you do this?

GTM leaders: any agent vendor that won't give you run logs and agent traces is a supply chain risk, and it will cost you valuation at exit.

A lot of agent vendors are playing the same game right now.

Build agents in our platform, in our UI.

No, you can't see the run history.

No, you can't see the tool calls, or how the skills you gave us are being used.

No, you can't build from where you already work, in Claude or ChatGPT.

That's frustrating. It will get painful

1. The vendor is learning from your data

Your vendor is almost certainly running evals (tests of whether the agent did the job: write the email, research the account, summarize the call) and collecting telemetry (logs of everything the agent did: its reasoning, MCP calls, skills used) in tools like Braintrust, LangChain or Langfuse.

Put those together with your data and they can see what works, change how the agents behave, and add instructions in the background you never see.

Most contracts only cover training models on your data.

Learning from your traces ends up in the same place: what your team taught the agent becomes the vendor's product.

2. At renewal, they hold what they learned hostage

"We understand you want to switch, but the agents have learned your company pretty well, so expect a performance drop."

Some of that is fair. Vendors should get better across all their customers, and I don't need their system prompts or every reasoning trace.

But if you have NONE of the data, you paid them to train their agent on your business, and now it's leverage against you.

3. Your company is worth less at exit

Companies are increasingly valued for the knowledge and data inside them.

Google just paid $10M in bankruptcy court for Spirit Airlines' emails, Teams chats and code, to train AI.

Your GTM work is no different.

Expect acquirers to start asking for agent traces and evals in diligence.

If your playbooks live as invisible prompts in a vendor's cloud, you own less of your GTM and the company is worth less.

What you can do today

1. Ask every agent vendor what telemetry and session logs you can export, and whether you can build agents, skills and workflows over MCP or CLI. Your own workspace has more context than their app.

2. Have engineering turn on log exports (OTEL where it's supported) from Claude, ChatGPT, or whatever your team uses. If you don't collect it, it's gone.

3. For your top use cases, paste 10 good and 10 bad examples into a chat connected to your tools and GTM context. Do 10 more with the agent, then have it write a skill. That's what your vendors do at scale, and they don't know your function or your company like you do. It's doable

4. Bring your CFO and exec team in. You WILL be judged on your agent infrastructure.

5. Favor vendors that give you your data. So far I'm bullish on @clay, @ZoomInfo, Day AI and @MonacoGTM. I'm sure there are others.

No, I was there when GTM engineering was getting built and defined as a term, and tool sprawl was not the problem.

The reason GTM engineering was created was simple: @kareemamin wanted to bring the power of programming to GTM.

@kylewill was doing this before Clay at @stripe and others, but at scale Product managers didn't really get GTM enough, and engineering didn't have the bandwidth to apply that kind of programmatic logic to GTM before that point

So GTM engineering + @clay was necessary to automate the flows that actually worked in specific GTM motions.

Now that same approach is possible in other platforms.

I agree an end to end platform is necessary... but if you REALLY look hard at it it's a problem that truly requires a platform that unlocks the creativity of users while holding a point of view on how work should be done.

It also requires a deep understanding of how to do each function extremely well, which is very hard to do.

At this point @clay is best positioned to be the Crowdstrike/ServiceNow of GTM, though they have quite a ways to go to catch all the edge cases and other channels, build out a knowledge graph + ingestion pipeline, and build an always on GTM agent harness.

Aurasell seems like a great solution to replace Gong for CI + SEP and some of Clay's functionality.

I think the value pyramid is genius, the deal coaching is effective as well, and it does a better job than most encoding sales best practices.

For a late adopting complex B2B sales team, it's definitely a better solution than Gong + self implementing clay without a partner.

But the agents likely aren't powerful or flexible enough for power users, and there's no way for you to match the combined data capabilities of a data marketplace for all sales motions.

The dialer will DEFINITELY get outcompeted by Titanx, or even likely by nooks, and it looks like you don't have multiple inboxes.

And it doesn't look like you have a skill library, the ability to access agent logs via MCP/CLI, a reference library, or a knowledge. So the quality of the outputs from a content perspective will likely not be as good, and the agents may not be aware of nuances in GTM motion and positioing.

I could go on about forms, website visitors, events, social engagement and other points. And I haven't even talked about MCP/CLI.

My point is, being a true all in one that's a market leader is insanely, psychotically difficult, and unlike other markets, in GTM, it makes financial sense to always pick the best tool for the job and integrate it.

The only way to actually be the all in one in GTM will be to integrate with everything, do everything, and essentially let your users built custom apps.

I sort of agree, and I like the agentic review of every email... but I think account context + detailed positioning + humans is how you reall win

We got a customer 1 meeting per 80 emails + 1 meeting every 35 dials by building incredible account research that we combined with detailed personas, value props, and specific campaign templates.

And this was a TAL with 300 accounts TOTAL

@titanXio made the connect rates possible, the activity volume overall was pretty low, and we had a very sharp outbounder doing the work.

But we never could get the model to fully grasp the nuances otherwise.

Might be possible with a knowledge graph. We'll see.

Love this. This is why we're shifting to a bootcamp motion where we take an MVP agent + predictive scoring model live with customer context in <3 weeks.