This post was originally published on LinkedIn.

Most GTM teams (even those w/ Claude Code) are stuck at Level 2 (AI Assistant)

Level 2 can get you some incredible gains. But it's still mostly the old way of working but with AI helping you along the way. You're still the bottleneck.

For example, Account Research:

Level 1 'The AI Thought Partner':

You paste account data into Claude → chat back and forth → copy findings → paste into CRM. Useful but not transformative by any means.

Level 2 'The AI Assistant':

Rep connects Claude Cowork to CRM → runs an Account research Skill where Claude pulls CRM data and does web research → builds a brief → rep reviews and pushes output back into CRM → pushes output into sales email tool. Very useful, saves a lot of time, but still needs a rep to get the process going and guide.

Level 3 'Production Grade AI Automation':

Automated workflow triggers automatically when Account is created or handed off. AI pulls CRM data + conducts web research in parallel → compiles brief → delivers to Slack and CRM fields → builds emails in sales tools. This is where it gets very interesting. Zero friction for reps, they just need to pick up the ball and act.

Level 4 'The AI Agent':

Fully autonomous. AI is working in CRM/MAP/etc production environments. Decides what context matters, does deep research using Salesforce, web, win/loss data, call transcripts, intent signals. Delivers insights to the right person at the right time, then learns from outcomes and refines its approach (it self-improves).

MOVING UP THE LEVELS

To get from L2 → L3 you need the process automated in production.

To get from L3 → L4, the agent needs freedom to act non-deterministically (i.e., the agent has a guide and goal, not pre-built automation) AND the agent needs to be self-improving.

THE BIG UNANSWERED QUESTIONS

At Level 3 and 4, you need to architect a system and not just provide a tool or skill to your team.

So you need to answer multiple questions:

Build vs. buy? Security? Safeguards? Monitoring? Can we build this ourselves with Claude Code? Should we buy Clay and configure it? What's the best path to the agentic layer? Can we bet on a tool when everything is moving so fast?

The answers are different for every company depending on their stack, their data, their risk profile, and their team.

But most companies can't even get there because none of this builds itself and they are swamped in the day to day. So most companies will remain stuck at L2 for a while.