Skip to main content
Back to all posts
Claude Cowork

How Sprintt Uses Claude Cowork to Run Client Projects

Cowork is the collaboration surface we live in during engagements — shared agents, shared skills, shared context across the Sprintt team and the client team. Here's the actual playbook.

Ricardo Ramirez

Ricardo Ramirez

Founder · Sprintt

March 14, 20268 min read
Claude CoworkWorkflowsConsulting

People ask us how we deliver Sprintt engagements at the speed we do with the team size we have. The honest answer is that we stopped running client projects through email and Slack. We run them through Claude Cowork.

Cowork is Anthropic's collaborative agent surface — the place where multiple people share agents, skills, and context in a single workspace. It looks, at first glance, like a fancier version of a Claude chat. It isn't. Used correctly, it is the central nervous system of a modern consulting engagement.

Here is the playbook we use, honestly and in specifics, so you can adopt it for your own work.

Why Cowork over alternatives

The consulting world has been stuck in a document-and-meeting paradigm for thirty years. The outputs of work are decks, memos, slack threads, emails. The process of work is meetings, standups, reviews. Both are optimized for a world where AI is a thing you mention in passing, not a thing you build with.

Cowork changes the paradigm in one specific way: it makes the AI the persistent collaborator, visible to the whole team. Every session is shared. Every skill is shared. Every context window includes what the whole team has been doing.

Slack is where people talk. Cowork is where work happens.

The workspace setup

Every client engagement at Sprintt starts with a fresh Cowork workspace. The setup takes about thirty minutes and is roughly the same every time. The repeatability is the point.

Workspace naming and access. Named for the engagement. The client's relevant team members are added on day one. Access is scoped by project.

Skills installed. We install our standard engagement skills — discovery prep, financial review, SOW generation, client onboarding, proposal drafting. The client sees them and understands that these are our repeatable motions, not ad-hoc magic.

Client-specific skill stubs. We add two or three skill stubs specific to the client: their style guide, their security review process, their deployment gates. These are filled in progressively as we learn.

Shared context documents. A living document with the engagement overview, stakeholders, timeline, outcome metric. Every session starts by referencing it.

Integrations. Depending on the engagement: the client's Linear or Jira, their GitHub or GitLab, their documentation source. Cowork becomes the place where agents can reach into the client's world.

Thirty minutes, every time, no exceptions. The setup discipline is what makes the engagement repeatable.

The three rhythms of an engagement

Every Sprintt engagement runs on three rhythms, and Cowork is the home for all three.

Rhythm 1: The daily working session

Not a meeting. A session. The builders on both sides — us and the client — open Cowork in the morning and work alongside each other through shared agents.

Concrete example, from a recent engagement: we were building an AI-powered ticket routing system for a support team. Every morning, the Sprintt engineer and the client's tech lead would open a shared Cowork session, review yesterday's work, brief an agent on the day's tasks, and let it run while they did other work. They'd check in at midday and end-of-day to review what the agent had produced.

This cadence collapses the traditional "async work, sync meeting" loop into a single continuous stream. You don't "write up a status report" because the status is already visible. You don't "schedule a review meeting" because the review happens in the thread.

Rhythm 2: The weekly outcome sync

Once a week, the senior stakeholders on both sides — not the builders, the decision-makers — meet for an outcome sync. Thirty minutes, tops.

Cowork is pre-populated before the sync with a report generated by a "weekly review" skill. The report pulls from the work done that week, the open risks, the decisions needed. It's never blank. It's never a surprise.

The meeting is entirely about decisions. "Here are the three things we need to decide this week." Not status updates. Not recaps. Decisions, made, captured.

After the meeting, a summary goes back into Cowork as part of the shared context. The next session's agents know what was decided.

Rhythm 3: The monthly phase review

At the end of each phase (every 4–6 weeks for most engagements), we do a more substantive review — what was delivered, what was measured, what's next.

The artifact for this review is generated by a skill that pulls from the full engagement history. It is a proper document — the kind of thing that would historically take a consultant two days to write. It takes about twenty minutes. It's also more accurate, because it's drawing from the actual record rather than the consultant's memory.

The patterns that matter

Below are the patterns that separate Cowork-as-a-chat-tool from Cowork-as-a-production-system.

Treat the workspace as the canonical record

The most important discipline: Cowork is the truth. If a decision was made in Slack but not captured in Cowork, it didn't happen. If a document was emailed but not linked in Cowork, it doesn't exist for the project.

This sounds severe. It is. It's also the only way to keep a multi-person, multi-week engagement from devolving into a dozen half-remembered threads across five tools.

Concretely: at the end of every Slack conversation that made a real decision, the thread gets summarized into Cowork. Every external doc that matters gets linked. The rule is: if you can't find it from Cowork, it isn't part of the project.

Use skills as the team's operating system

We have a skill for every recurring motion in an engagement: discovery calls, invoicing, weekly reviews, security reviews, onboarding a new client team member. Each skill is a file. Each file is version-controlled. Each one captures the "how we do it" so no one has to remember.

The leverage compounds. Our seventh client onboarding is better than our first, because the skill has been edited seven times. Every mistake we made becomes an improvement in the skill. Every pattern we discovered gets encoded.

This is the biggest shift from traditional consulting: knowledge doesn't live in people's heads. It lives in skills. When a new Sprintt team member joins, they don't have to learn our tribal wisdom. They open Cowork and use the skills.

Agent memory matters more than you think

Within a Cowork workspace, agents accumulate memory — the key facts about the engagement that persist across sessions. This is where most teams drop the ball.

The right posture: treat agent memory like a living client dossier. When you learn the client uses Kubernetes on GCP, save it. When you learn their QA lead's name is Mike and he prefers bullet-point reports, save it. When you learn a particular API is deprecated internally, save it.

On day one, agents know almost nothing. By week three, they know what it takes us years to fully train a junior consultant on. That's compounded leverage. The team becomes smarter together.

Keep context discoverable, not searchable

Slack is searchable but not discoverable — you find something only if you remember it exists. Cowork should be discoverable: when an agent starts a new session, it has the relevant context loaded, whether or not the operator asked for it.

The mechanism is the workspace's shared context documents plus agent memory. Together they mean every new session starts partway up the learning curve, not at the bottom.

Delegate summarization to skills

Every engagement produces a lot of noise. The way we manage it: we have a suite of summarization skills that run on a schedule.

  • /daily-roundup pulls together what the team did today, flags open items, pre-fills tomorrow's brief.
  • /decision-log scans the workspace for decisions made, and records them into a durable log.
  • /risk-register scans for things that look like risks or concerns, and adds them to a prioritized list.

These skills are the scaffolding. They run automatically or on-demand. They mean nobody on the team has to remember to write the status report — the status report writes itself.

What goes wrong, and how we fix it

"The client's team isn't using Cowork." This is the #1 failure mode. If only Sprintt is in Cowork and the client is still in email and Slack, the compounding doesn't work. Fix: make Cowork adoption part of the engagement. We include it in the SOW. We run a 30-minute onboarding for the client team. We follow up when we see someone still emailing updates.

"The workspace is getting noisy." Without discipline, Cowork becomes another cluttered channel. Fix: archive finished threads. Use the pinned context doc to point to what matters. Run weekly decluttering — we have a skill for this.

"An agent gave wrong information that the team acted on." Rare, but it happens. Fix: require human review before any customer-facing or externally-shipped output. Log every decision so you can trace what went wrong. Update the relevant skill or memory so it doesn't happen again.

"The client is worried about their data in Cowork." Real concern. Cowork has enterprise-grade controls for this, and we walk through them in the contracting phase. Workspace data is scoped per-workspace; the client owns their tenancy; retention is configurable. Clients who understand this stop worrying.

The compounded effect

The reason Cowork works for us — not just as a nice tool, but as a strategic advantage — is that the leverage compounds per engagement.

  • Engagement 1: the basic skills are in place. Decent outcomes.
  • Engagement 5: skills have been iterated. Each one is measurably better. The engagement runs faster.
  • Engagement 20: the skills cover most of the recurring work. New engagements start with a mature playbook. The builders are free to focus on the novel problems.

This is the multiplier we can point to when asked how a small firm delivers enterprise work at the pace we do. The team is not working longer hours. The team is not larger. The team is running on better infrastructure — and the infrastructure is Cowork, with skills as its operating language.

If you run a consultancy, an agency, or any team that delivers recurring client work: you should be running this way. The alternative is being beaten by teams who are. Which, right now, is not a hypothetical.


Sprintt engages with new clients through a discovery call where we walk you through our Cowork-based delivery model in thirty minutes. If you're running client engagements the old way and wondering how firms are shipping faster, book a call.

Ricardo Ramirez

Written by

Ricardo Ramirez

Founder of Sprintt. Product leader, practitioner, and operator — not an academic or a theorist. Writes about the gap between AI strategy and shipped production systems, because closing that gap is the only thing Sprintt does.

Book a 30-min call

Ready to ship?

Stop planning.
Start shipping.

30 minutes. No pitch deck. A direct conversation about where AI can drive the most impact for your organization.

Book a strategy call