At Rumble Fish, a software development agency, we live in an industry defined by constant change. Adapt or become irrelevant - that's the reality we face every day. When AI coding assistants emerged, we were early adopters. But what surprised us wasn't Claude's ability to write code. It was everything else.

The Work That Couldn't Be Automated

Every agency has tasks that seem immune to automation: analyzing datasets to find patterns, correlating information from multiple sources, or making sense of unstructured data. This work requires judgment, context, and the ability to handle exceptions. It's why we always assumed it needed humans.

We were wrong.

A Different Kind of Intelligence

Claude doesn't just follow scripts. Its reasoning capabilities allow it to adapt when things don't go as expected. Feed it messy data, and it figures out what you meant. Throw it an edge case, and it works through it logically. You could almost call it creative - not in the artistic sense, but in the problem-solving sense.

We've built repositories that define skills Claude can use. These aren't rigid automation pipelines. They're flexible frameworks where Claude applies judgment within boundaries we set. The results have been remarkable.

The New Division of Labor

The efficiency gain isn't about replacing humans. It's about changing what humans do.

Before: A team member spends hours on repetitive analysis, losing focus, getting tired, and making mistakes on the mundane cases while staying sharp for the interesting ones.

Now: Claude handles the repetitive work with unwavering consistency. It never gets bored. It maintains the same level of attention on case number 500 as it did on case number 1. And when it hits those edge cases requiring genuine creativity? It still delivers.

Humans review results, refine prompts, and handle the truly novel situations. It's a better use of everyone's capabilities.

Claude in Practice

Lead Generation

This use case was put together in just a few days by some of our engineers as part of their AI training. Historically, our marketing and sales team spent considerable time scouring the web for potential clients. It's such a time and labor-intensive task that it almost felt like a waste of effort.

We built a Claude repository with skills that scrape publicly available job boards and identify relevant opportunities. This goes far beyond what a traditional search query could do. We're looking for companies that offer remote work, are open to hiring an agency, and need expertise in technologies we specialize in: blockchain, ZKP, AWS, and AI. Claude turned out to be remarkably good at understanding job posts and rating them according to the rules we defined.

The result? We can now process a flood of information that was previously too massive to handle manually. This has opened an entirely new sales channel for our agency.

Cloud Deployment for Team Collaboration

Most people encounter Claude as an IDE extension or a standalone desktop application. But there's another way: running the Claude SDK inside a Docker container.

This allows us to deploy Claude as an ECS task on AWS, which unlocks something powerful for team collaboration. Multiple team members can share the same knowledge base and continuously update the skills and rules defined in the repository. When someone refines a prompt or adds a new capability, everyone benefits immediately.

The result is a tool that effectively learns from the humans using it. Each improvement compounds, and the system gets smarter over time - not through model updates, but through collective human insight captured in the repository.

Find out more about AI-powered enterprise knowledge management solutions.

Looking Forward

We believe this represents a fundamental shift in how knowledge work gets done. Not automation in the traditional sense, not rigid rules and brittle scripts. Something more flexible. Something that can think. The agencies and teams that figure this out will operate at a different level of efficiency. The ones that don't will find themselves wondering how their competitors move so fast.

We're still learning what's possible. But one thing is clear: Claude isn't just a coding tool. It's a thinking tool. And that changes everything.