Anthropic released Claude Cowork about a week ago. My initial reaction was skeptical. The product was reportedly built in a week and a half using Claude Code, and my first impression was that it felt rushed. It didn't seem to deliver much beyond what was already possible.
I was wrong.
After actually using it and connecting it to the tools that run my consulting practice, my view has completely reversed. The results have been striking enough that I wanted to share what I've built and what I've learned.
Cowork allows Claude to connect to external systems through the Model Context Protocol, or MCP. I've hooked it up to three systems: Notion, Fireflies, and a custom MCP I built for GadociConsulting.com itself. The result is a single conversation that can read from and write to all three.
What Cowork Adds Beyond the Web
To be clear, much of what I'm describing is available through Claude's web interface. You can connect MCPs, query external systems, and orchestrate actions across tools. But Cowork introduces a few things that change the experience meaningfully.
First, it uses the same planning and reasoning architecture that powers Claude Code. If you've used Claude Code for software development, you know there's something different about how it approaches complex, multi-step tasks. It thinks before it acts, breaks problems into stages, and maintains coherence across longer workflows. Cowork brings that same approach to non-coding work, and the quality improvement over the standard web experience is noticeable.
Second, Cowork can work with local files. This matters more than I expected. I have documents on my desktop that aren't published anywhere: draft proposals, client notes, strategy documents, half-finished thinking. With Cowork, I can point Claude at those files directly. "Read this draft proposal and compare it to what we discussed in yesterday's call" becomes possible without uploading anything or copy-pasting content. The local file system becomes part of the context.
These aren't dramatic differences, but they add up. The planning architecture means fewer misfires on complex requests. Local file access means my unpublished work can inform Claude's responses without extra steps.
The Notion Hub
My Notion workspace is the operational backbone of Gadoci Consulting. It's not just a task list. It holds eight interconnected databases: Clients, Projects, People, Meetings, Tasks, Time Tracking, Notes, and 90-Day Plans. These databases link to each other, so a meeting record connects to the client, the project, and the attendees. A time entry links to the team member, the client, and the project. Notes cross-reference people, projects, meetings, and tasks.
This relational structure is powerful, but it also creates overhead. Creating a meeting record used to mean navigating to the right database, filling in multiple relation fields, and making sure everything was properly linked. Now I can say "log my meeting this morning, 45 minutes, link it to the governance project and add the attendees." Claude handles the lookup and linking automatically.
The same applies to time tracking, task creation, and note-taking. The friction of maintaining proper relationships across databases disappears when I can describe what I want in plain language.
From Transcript to Action
Fireflies captures my meeting transcripts automatically. That's useful on its own, but the real value comes from what happens next. I can now open Claude and say "pull my transcript from this morning's call and create action items in Notion." Claude reads the transcript, extracts the commitments and decisions, and creates properly tagged tasks in my Notion workspace with the right client and project relationships. What used to take fifteen minutes of copy-paste and manual linking now takes one sentence.
The website integration extends this further. I can turn a meeting transcript into a knowledge base article with a single request. If a client conversation surfaces an insight worth capturing, I ask Claude to draft an internal article summarizing the key points. If it's something worth publishing publicly, I can have Claude create a draft article on GadociConsulting.com and set it to review status before I polish and publish.
Voice as the Interface
Here's the part that feels like the future: I'm not typing most of this. I use Wispr Flow, a voice dictation tool that works across every application on my Mac. I speak naturally, and it converts my voice to clean, formatted text in whatever app has focus. It's about 4x faster than typing.
Combined with Claude Cowork, this changes the entire interaction model. I can walk around my office, thinking out loud, and direct Claude through voice. "What's on my plate this week?" "Log 2 hours on the governance documentation." "Summarize yesterday's transcript and create a follow-up task for next Tuesday." No keyboard. No clicking through interfaces. Just speech to action.
This is what the future of knowledge work looks like. The bottleneck has always been the translation layer between thought and execution. Typing is slow. Navigating interfaces is slow. Context-switching between apps is slow. When you can simply speak your intent and have it execute across multiple systems, the friction approaches zero.
The Compounding Value of Context
The real power isn't any single action. It's that Claude has context across all three systems simultaneously.
When I ask "what should I focus on today?", the answer draws from my Notion tasks, recent meeting transcripts, and any pending content deadlines. When I'm preparing for a client call, I can ask Claude to surface relevant notes from past meetings and any outstanding action items in one response.
This kind of cross-system awareness is new. Each tool used to know only its own corner. Now I have an interface that understands my practice as a whole.
The Technical Pieces
For those interested in the implementation: Notion and Fireflies integrations use their official APIs wrapped in MCP connectors. The GadociConsulting.com integration is a custom MCP I built that exposes my site's content management, user management, and analytics through a structured interface. Claude can create articles, manage prospects, track learning content, and pull analytics, all through natural language requests.
The MCP specification is straightforward if you're comfortable with APIs. The harder part is deciding what capabilities to expose and how to structure the interactions so they feel natural in conversation.
What I've Learned
Three observations after a week of working this way.
First, the reduction in context-switching is more valuable than I expected. Staying in one interface, thinking in one mode, preserves a kind of focus that's hard to quantify but easy to feel.
Second, the barrier to documentation drops dramatically. Meeting insights that might have stayed buried in transcripts now get captured because it costs nothing but a sentence.
Third, this changes what's worth building. I'm rethinking how my practice runs because the friction equation has changed. Processes that seemed too expensive to formalize are now cheap. That opens up possibilities I hadn't considered.
I went in skeptical and came out a convert. If you're curious about Cowork, my advice is to actually use it before judging it. The product punches above what its quick development timeline might suggest.