HubSpot product and marketing leaders on scaling Claude Cowork

HubSpot is an agentic customer platform that helps marketing, sales, and customer service teams drive growth. Both product and marketing teams at HubSpot are scaling Claude Cowork across teams. We spoke with nine HubSpot leaders to walk through how their teams are using Cowork in practice. 

How did your teams first start using Claude Cowork, and what were you hoping it would help with?

Chloe Tambe, Director, AI Transformation: The HubSpot Marketing team didn't come to Cowork as strangers to Claude. We were already extensive Claude users: individual chats, shared projects, artifacts, integrations woven into daily work. So when Cowork came along, the question wasn't "should we try this?" It was "how fast can we go further?"

Cowork's automation really appeals to the Marketing Team. We have hundreds of marketers with very different jobs, and the administrative load is real. Cowork gives us the ability to schedule and run tasks even when people are offline, taking busywork off their plates so they can focus on actually being marketers. A thought partner that works alongside the team, not just on demand, is a meaningful shift.

Kathie Lui, Product Manager, Central AI: The specific pain point was connector sprawl. As Claude Cowork rolled out across HubSpot, I was managing requests to add new tool integrations through ServiceNow requests and Asana tickets, while simultaneously fielding questions in Slack, tracking approvals in Google Sheets, and coordinating with our security review team. It was the same kind of work happening in four different places. Cowork didn't just help with one workflow. It was the first tool that could sit on top of all the others. For someone who collaborates across a lot of different teams and platforms, that continuity is a big shift.

Walk us through a specific Cowork workflow that's become part of a regular routine for your team.

Sejal Parikh, Staff Product Manager: The biggest shift was thinking about skills as team assets, not personal ones. If I've built a workflow that saves me 30 minutes a day, my whole team probably needs it too. We have a team-level context repository where customer insights, decisions, data analysis, RFCs, and key Slack threads all live in one place, so any of us, or any agent, can pull from the same source instead of redoing the research. Run agents in parallel for the heavy lifting, it saves time and keeps your own context window clean. We also share skills for repeatable tasks so anyone on the team can use them.

Tim B., Principal AI Technical Specialist: I needed to build an internal AI enablement hub for our blog team: walkthroughs, prompt frameworks, Loom tutorials, polished UI. The kind of thing that normally takes weeks of wrangling across HubSpot's scattered doc infrastructure. Instead I built it as a Claude artifact on our enterprise plan, gated to HubSpot by default, using skills I built and iterated on in a single Cowork session. The flow: record a Loom, paste the transcript, a skill I built converts it into a formatted walkthrough and links the video automatically. All my reference context lived in a local folder Cowork could reach directly. No upload loops, no re-establishing context. Full V1 in about 3 hours. Updates now take as long as it takes to record a Loom.

Oscar Estrada, Senior AI Creative Technologist: I built a dashboard for my team that tracks all the AI video tools we're piloting. That dashboard is a Claude artifact that uses a JSON file on the back end, containing the entries for the various tools, with all the metadata that populates the dashboard. I also have a Claude skill that’s in charge of updating that dashboard periodically with new information pulled from notes, meetings, and other web sources. The dashboard also connects users to trainings and workflows. Leveraging Cowork's file system access and sub-agents has been key in helping me keep the JSON file up to date.

When you're using Cowork to generate first drafts of documents or plans, what do you provide as input, and how usable is the initial output versus what you end up revising?

Daniel Jacobson, Director of Product Management, Developer Products: Most of my first-draft work in Cowork is strategy memos and decision docs. The input is usually a short brief plus pointers: a Slack thread or a meeting transcript. I wrote a memo-writer skill that captures my voice and decision frameworks so I don't have to re-prompt that every time. The first output is usually 60 to 70% of the final shape. It pulls in context from connectors I didn't have to paste. Most of my revision is sharpening claims, cutting hedge language, and adding the spiky take only I can write. The unlock isn't "AI writes my doc." It's that the time from idea to reviewable draft collapsed from two hours to twenty minutes.

Estrada, Senior AI Creative Technologist: When I'm building docs, I provide multiple inputs, including Slack messages, documents, Asana tasks, and anything that can be connected to Claude. What I find most helpful is having Claude list its assumptions and ask me any questions it needs before building the doc. Cowork also makes it easy to edit docs without having to copy and paste from the artifact to a separate file and back. If I'm creating a type of document multiple times, I usually create a skill for it. That gets me over 80% of the way there.

Maggie Flinn, Lead Program Manager: I give Cowork the raw materials and a clear ask. For a recent enablement roadmap, I described the parameters: quarterly goals, monthly milestones, focus on user engagement across our AI tools, plus context about team structure, adoption data, and governance needs. What came back was a six-pillar roadmap with phased goals across the full year, specific deliverables on timelines, and a logical progression from foundation to scale to optimization. That structural thinking is what saves the most time. Organizing parallel workstreams into a coherent, sequenced plan is what would normally take me hours. Instead I was refining a draft that already had a point of view.

For executive decks, same pattern. I upload raw analysis from a tool rollout and say "8 slides, cut the granular data, spotlight adoption momentum, end with three asks." I get back a narrative arc with speaker notes and strategically placed data points. The structure and framing are usually 80 to 90% usable. Where I do rework is the stuff that requires my context, like pressure-testing whether the asks are realistic given actual resourcing constraints.

Iteration matters. I don't take the first output and run. I'll say "flag where I'm under-resourced" or "reframe this as an ask instead of a status update." Each round sharpens it. The workflow is now less "generate and hope" and more fast collaborator who gets me to a strong starting point in minutes. Now I spend more time on judgment calls.

What surprised you most about how Cowork has spread inside HubSpot?

Kolin Koehl, Sr. Director of Product Management, GM of Service Hub: How quickly it clicked for PMs. It started as a pilot and immediately turned into us onboarding our entire PM org. We have a Slack channel for PMs to share their AI experiments, and after rolling out Cowork, PMs filled it with demos of how they set up their Cowork brain and what workflows they're executing. The learning was contagious. Cowork is not just a tool for PMs, it's a new way to get product work done.

How did this work look before Cowork, in terms of time and people involved?

Tim B., Principal AI Technical Specialist: Before Cowork, the content wasn't hard, the infrastructure was. At HubSpot, building something polished and gated for non-technical writers means filing requests, learning new tools, or producing something that looks DIY. The bigger hidden cost was context: before using Claude Cowork, I'd be manually uploading docs and re-establishing context every session. Now everything lives in one local folder for the entire build. One person, one session, three hours with rave feedback the same week. The quality was way better than it would have been without Cowork. A multi-week project with multiple teams involved is now typically reduced down to a few hours. 

Jon Blackwell, Manager, Business Systems: Writing a Business Requirements Document used to be the part of every project that nobody looked forward to. The process was the same every time, and not in a good way. It involved reviewing handwritten notes, comparing versions across the team, and coordinating reviews before anything could be finalized. It was days or weeks of overhead before the real work could begin.

Now, the BRD shows up to the kickoff meeting before we do. Using Cowork, we developed custom document-formatting skills built specifically for how our team works. By the time we sit down for that first meeting, a first draft is already built. As the project moves forward, meeting transcripts and new inputs get added to the session, and the document evolves with them.

What advice would you give to another team thinking about getting started with Cowork?

Jacobson, Director of Product Management, Developer Products: Three things made this work for our team. First, build skills, not prompts. The first time I asked Claude to write a memo, the output was generic. The hundredth time, with a memo-writer skill that captured my voice and decision frameworks, it was 70% there. Skills compound. Prompts don't. Second, connect everything. The leverage isn't the model. It's the model with access to your Slack, transcripts, Calendar, issue tracker, and Gmail. A daily briefing that pulls live data is worth ten generic productivity tools. Wire up connectors before you spend time prompting. 

Third, treat your personal and team knowledge bases as architecture. Information scaffolding improvements often compound as much as, or more than, model improvements. Spend the time investing in your information scaffolding and it will pay dividends. Pitfall: don't let it stay generic. Week one feels magical, week five feels disappointing if you didn't invest in scaffolding. Skills + connectors + structured memory turn it from a chatbot into a thinking partner.

Blackwell, Manager, Business Systems: Don't skip the setup. Invest time upfront to establish your connections, identify relevant pre-built skills, and configure settings that fit your workflow. Build out a thorough personal context file. The more context Claude has, the more useful its outputs will be. Think of it less like configuring software and more like onboarding a new teammate. An early win: start with a Mission Control Dashboard. Once your connections are configured (Google Drive, Asana, Slack, calendar), Claude can pull context from all of them into one place. It's a low-effort, high-visibility win that makes the power of connected context tangible, and tends to be the moment teams go from curious to convinced.

Parikh, Staff Product Manager: The skill I'd never give up is my daily reflection. It pulls from Gmail, Calendar, Fellow, Slack, Google Docs, and Linear and hands me a short EOD recap with top follow-ups already drafted in my voice. A writing style guide skill keeps every draft sounding like me. My context lives in my Obsidian vault, so I move between Claude Code and Cowork without losing anything.

Koehl, Sr. Director of Product Management, GM of Service Hub: Feed it your actual work. Tell it to run your weekly competitive analysis and ask it to find pain points in customer interviews you may have missed. If you only use Cowork for five-minute chores, you'll never feel the step-change in leverage.

‍