A few months ago, my workflow for staying current on the product looked like one of two scenarios.
Scenario one: I'd open Super, our AI work assistant, and ask "what shipped recently? What's coming up I should prepare for?" Super would surface links to source docs in Linear or Slite. I'd read through them, and my marketing brain would start spinning: does this need sales enablement? A new one-pager? An updated website section? The information was there, but scattered across tools, and the path from "what shipped" to "what marketing needs to do" was entirely manual.
Scenario two: a product or engineering lead messages me, "hey, this is launching this week, can you run go-to-market on it?" Now I'm scrambling to get up to speed on a feature I'm hearing about for the first time, figure out what GTM motions it needs, and execute, all in the same week.
Both scenarios share the same problem. Marketing is downstream of where the product gets built.
Even when I started using AI to help, the workflow had its own version of this problem. I had Claude Desktop projects set up for each product line, each with its own attached files: narrative decks, blogs about how the product works, competitive positioning docs. But with AI letting engineers ship faster than ever, so much can change within the product in a single week. So much can change in the market in terms of competitor positioning within a single day. That meant I needed to update the files in my projects weekly, ideally daily, just to keep the AI's context accurate. And even then, my chats were siloed. The rest of the company couldn't see what I was building or reuse any of it.
So I moved to where the builders live.
What Happens When a Marketer Works in a Repo
The engineering repo at Beyond Identity has everything: product specs, roadmap plans, release notes, customer call transcripts, competitive analysis, and the latest feature documentation. It's the single source of truth because that's where product and engineering already work. They don't export their knowledge to a marketing wiki. They build here.
The problem: repos are not built for marketers. Git commands, file structures, markdown, pull requests. None of this is intuitive if you've spent your career in Google Docs and Canva. I was already using AI heavily with Claude Desktop projects, but the shift to Claude Code in the repo changed the game entirely.
I use Claude Code every day now. It reads the repo, understands the product context, and helps me produce marketing output directly from the source. This means no more Slack pings, no "can you send me the latest?", and no version confusion, because the repo is the source of truth and Claude Code is what makes it accessible to someone who isn't an engineer.
Skills Turn Tribal Knowledge into Repeatable Systems
Raw access to the repo isn't enough, just like raw access to Claude Desktop wasn't enough on its own. A marketer staring at product spec markdown files still needs to know what to do with them.
That's where skills come in. Skills are instructions I've built inside Claude Code that define exactly what the output should look like. They enforce our brand's tone of voice. They remove AI slop (em dashes, structural anti-patterns, formulaic phrases that scream "a robot wrote this"). They ensure every piece of content tells a clear story. They even pull from external resources and documented best practices.
The output I produce this way:
- Nurture email sequences that reference the latest shipped features
- SDR and sales outreach built on the pitch that's actually working on customer calls right now
- LinkedIn thought leadership posts grounded in real product capabilities, not vague positioning
- Digital advertising copy that matches what the product actually does today
- One-pagers drafted from the same specs engineering uses
I could produce some of this with Claude Desktop projects before, but each project was its own silo with its own files that I had to keep current. The repo flips that: the source of truth is maintained by the people building the product, and a single Claude Code prompt pulls it all together for whatever marketing output I need.
Building Marketing Infrastructure Without Writing Code
The time savings alone would make this worth it, but the real shift is what became possible that wasn't before, even when I was already using AI.
I built an automated pipeline that connects our product's sign-up API directly to HubSpot. Every new product sign-up triggers an automated nurture sequence. The system enriches each sign-up with company and contact data pulled from the web. All of it flows into a single pipeline tracker that the CEO and executive team can check anytime for the latest on product usage, sign-up demographics, and engagement data.
Before AI, this would have been a project request to engineering, a ticket in the backlog, and weeks of waiting. Instead, I built it by working with Claude Code in the repo, describing what I needed, iterating on the approach, and shipping it.
What the Numbers Look Like
The hours saved per week are significant. But the bigger metric is what I can now do that I couldn't before: build systems, automate workflows, and produce content at the speed the product ships.
The Entire Marketing Team Went AI-First
This isn't just my workflow. The entire Beyond Identity marketing team now works AI-first, with the repo as our shared workspace, Claude Code as our daily tool, and skills encoding our standards so output is consistent regardless of who's producing it.
The pattern is simple: put the source of truth where the builders already work, use AI to make it accessible to everyone else, and encode your quality standards so the output doesn't depend on who's in the seat.
When Every Team Is AI-Native, Security Can't Be Optional
Working in Claude Code means sharing API keys. HubSpot API keys for the nurture automation. Ceros product API keys for the sign-up pipeline. These are real credentials with real access to production systems.
I work in cybersecurity, so I generally know which actions are safe and which aren't. But that's not the norm for marketers, or any non-security role. Security's oldest lesson applies here: don't build your defenses around one person doing the right thing, because eventually someone won't.
Full disclosure: I work at Beyond Identity, and our product Ceros is built for exactly this problem.
When I launch Claude Code through Ceros, my API keys never touch my machine. They stay locked in a hardware security module (HSM) in our cloud and get injected into requests at runtime by the cloud proxy. Claude just sees a local connection on my machine, it has no idea where the real keys are. If my laptop were stolen tomorrow, the attacker wouldn't find any API keys to steal because they were never there to begin with. My device credentials (the signing keys that authenticate my requests) live inside the device's secure hardware and can't be extracted, even with root access.
I don't have to remember to handle credentials safely because the system enforces it for me. That's governance by infrastructure: security that works because of how the system is built, not because every individual on the team made the right call. Every action Claude Code takes is tied back to me, on my device, at a specific time, so if something goes wrong there's a full audit trail with actual proof, not a "probably Claude" story.
As more teams go AI-native (and they will), this kind of infrastructure becomes a requirement. The marketing team shouldn't need a security certification to use AI tools safely.
Where This Goes
The direction is clear. Marketing teams that treat AI as a daily tool, not an experiment, will produce better content, faster, from more accurate sources. The teams that work where the builders work, instead of waiting for builders to export their knowledge, will always be closer to the truth.
The repo-as-source-of-truth pattern scales to any function: sales enablement, customer success, technical writing, partner marketing. Anyone who needs to stay current on a fast-moving product benefits from being where the product is built.
The work now is making this accessible to every team, and the pieces are already there: skills that lower the barrier, AI tools like Claude Code that remove the technical prerequisite, and governance infrastructure like Ceros that makes it safe for non-technical users to work with real systems and real credentials without security becoming a bottleneck.
The builders moved fast. Now the rest of the company can too.