I have a new job. I've joined Inflection as CMO.

This comes from last week's news that Inflection acquired Keyplay.

My reasons for joining are typical: Team, market opportunity, mission that fits my strengths, exciting role, right timing. And for bonus points, I'm working for a CEO I've known and trusted for 15 years.

But my onboarding has been different.

Over the last 5 months I've become "extremely AI pilled" (as Ramp calls it).

So as I enter this new role, I have to go full AI-pilled and be an "AI-Native CMO."

I honestly don't know yet.

But I think the only way to figure it out is to take a pretty extreme posture and avoid doing a lot of the traditional onboarding steps. So that's what I did. I figure this will bring a lot of learnings that I can share here along the way.

To start, here's what I'm doing for a "full AI" CMO onboarding.

1.) No traditional 30/60/90. First week, I set up a Claude Code personal operating system. Local files organized into three buckets: context, work, system. Every project starts in Claude Code. Every decision and output flows back into context. The system gets smarter every day because the context compounds. Here's my simple architecture, not 1,000 folders:

2.) Built Claude skills WHILE I work. I didn’t just sit down and write skills. I started doing work, then build skills around that work. After a couple weeks I had skills to analyze pipeline data from Salesforce, search customer call transcripts for key positioning nuggets, surface content ideas, draft campaigns, and track team activity. I have a /morning skill that gives me a daily feed of what's happening internally (Slack, Notion, Calendar) and externally (calls, SFDC, LinkedIn). It's like a personalized briefing that knows my priorities.

3.) Launched a centralized "GTM Context" resource for the team. I helped create a central knowledge base that everyone's agents can draw from. Positioning docs, customer stories, call transcripts, ICP definitions, win-loss analysis. The idea is that any agent on the team (mine or anyone else's) has access to the same ground truth about our customers and market. We also create a quick start so that new people can have their Claude hook-up to the folders instantly.

4.) Hosted a live Claude Code setup session with the entire GTM team. I did a zoom with the entire GTM org (13 people) to share set-ups and help everyone get a similar baseline. For me it's Claude Code, VSCode, local folder structure, MCPs, APIs. This was not a polished demo. Just screen-sharing the messy reality and trading notes.

5.) Built a "Write with Claude" kit to help my team setup custom voice files & writing skills. I haven't found that AI can write for me, but Claude can be very helpful at two things: 1. Outlining and structuring my raw insights; 2. Editing flow and prose. That said, it took a lot of customization over the last few months to make Claude Code a decent writing assistant — I wanted my team to get that faster. The key was to have highly curate "voice" files that always get pulled into context for any writing task and specific skills that wrap those voice files. For me the core skills are /draft-content and /iterate-content.

Oh, and a new website. Not a AI project, but it's almost a meme for the new CMO to do a website redesign in his first month on the job. So I had to do this right away. The new site looks great IMO (inflection.io). If you agree, tell my boss it was a great idea.

It's just been a few weeks. Some honest observations:

Going from single-player to multi-player brings all the complexity. I was feeling great over the last months living in my Claude Code personal OS. I was able to modify this for my new role and company very quickly. By now trying to bring this to the whole team I see the biggest challenge. Once things go multi-player, it gets complex. The whole deployment, security, collaboration model is very early. Especially in GTM (I think engineering teams have figured out more). To start, we’re trying to avoid boiling the ocean, implementing some key legos, and then finding the right patterns for each situation.

Any effort on centralized “context engineering” is worthwhile, and compounds. AI agents are only as good as the context they have. The boring work of organizing and maintaining context is one of highest-leverage thing I've invested in so far — both for myself and the team. And as we created the central GTM Context system, others immediately started contributing and expanding it. This will compound fast.

Resist the urge to over-architect. When I first start in Claude Code a few month back, I over-built. My first instinct was to build a perfect system with clean hierarchies and comprehensive frameworks. That was wrong. The best approach is to do real work, notice what repeats, and build systems around actual patterns. The system emerges from the work, not the other way around.

I'll be sharing more as I go. I'm excited to figure out what works, what breaks, what surprises me in the process. I'm not planning to offer any sort of "copy paste replace your team comment for access" bag of markdown. But I am hoping to engage in the conversation deeply with other "AI pilled" GTM leaders and practitioners.

If you're a GTM leader leaning on AI, I'm curious what's pulling you in and what's holding you back?

I read all replies and would love to trade notes.