If you've ever used a 3D printer, you may recall the wondrous feeling when you first printed something you could have never sculpted or built yourself. Download a model file, load some plastic filament, push a button, and almost like magic, a three-dimensional object appears. But the result isn't polished and ready for mass production, and creating a novel shape requires more skills than just pushing a button. Interestingly, today's AI coding agents feel much the same way.
Since November, I have used Claude Code and Claude Opus 4.5 through a personal Claude Max account to extensively experiment with AI-assisted software development (I have also used OpenAI's Codex in a similar way, though not as frequently). Fifty projects later, I'll be frank: I have not had this much fun with a computer since I learned BASIC on my Apple II Plus when I was 9 years old. This opinion comes not as an endorsement but as personal experience: I voluntarily undertook this project, and I paid out of pocket for both OpenAI and Anthropic's premium AI plans.
Throughout my life, I have dabbled in programming as a utilitarian coder, writing small tools or scripts when needed. In my web development career, I wrote some small tools from scratch, but I primarily modified other people's code for my needs. Since 1990, I've programmed in BASIC, C, Visual Basic, PHP, ASP, Perl, Python, Ruby, MUSHcode, and some others. I am not an expert in any of these languages—I learned just enough to get the job done. I have developed my own hobby games over the years using BASIC, Torque Game Engine, and Godot, so I have some idea of what makes a good architecture for a modular program that can be expanded over time.
Caleb John (left), an investor with Pioneer Square Labs, and Lucas Dickey, a longtime entrepreneur, helped host the Claude Code Meetup in Seattle on Thursday. (GeekWire Photos / Taylor Soper)
Claude Code has become one of the hottest AI tools in recent months — and software engineers in Seattle are taking notice.
More than 150 techies packed the house at a Claude Code meetup event in Seattle on Thursday evening, eager to trade use cases and share how they’re using Anthropic’s fast-growing technology.
Claude Code is a specialized AI tool that acts like a supercharged pair-programmer for software developers. Interest in Claude Code has surged alongside improvements to Anthropic’s underlying models that let Claude handle longer, more complex workflows.
“The biggest thing is closing the feedback loop — it can take actions on its own and look at the results of those actions, and then take the next action,” explained Carly Rector, a product engineer at Pioneer Square Labs, the Seattle startup studio that organized Thursday’s event at Thinkspace.
Software development has emerged as the first profession to be thoroughly reshaped by large language models, as AI systems move beyond answering questions to actively doing the work. Last summer GeekWire reported on a similar event in Seattle focused on Cursor, another AI coding tool that developers described as a major productivity booster.
Claude Code is “one of a new generation of AI coding tools that represent a sudden capability leap in AI in the past month or so,” wrote Ethan Mollick, a Wharton professor and AI researcher, in a Jan. 7 blog post.
Mollick notes that these tools are better at self-correcting their own errors and now have “agentic harness” that helps them work around long-standing AI limitations, including context-window constraints that affect how much information models can remember.
On stage at Thursday’s event, Rector demoed an app that automatically fixed front-end bugs by having Claude Code control a browser. Johnny Leung, a software engineer at Stripe, said Claude Code has changed how he thinks about being a developer. “It’s kind of evolving the mentality from just writing code to becoming like an architect, almost like a product manager,” he said on stage during his demo.
Johnny Leung, a software engineer at Stripe, demos Claude Code and shows a tweet from Boris Cherny, the Anthropic engineering leader who created Claude Code.
R. Conner Howell, a software engineer in Seattle, showed how Claude Code can act as a personal cycling coach, querying performance data from databases and generating custom training plans — an example of the tool’s impact extending beyond traditional software development.
Earlier this week Anthropic — which is reportedly raising another $10 billion at a $350 billion valuation — released Claude Cowork, essentially Claude Code’s non-developer cousin that is built for everyday knowledge work instead of just programming. Anthropic on Friday expanded access to Cowork.
AI coding tools are energizing longtime software developers like Damon Cortesi, who co-founded Seattle startup Simply Measured in 2010 and is now an engineer at Airbnb. He said Thursday’s event was the first tech meetup he’s attended in more than five years.
“There’s no limit to what I can think about and put out there and actually make real,” he said.
In a post titled “How Claude Reset the AI Race,” New York Magazine columnist John Herrman noted the growing concern around coding automation and job displacement. “If you work in software development, the future feels incredibly uncertain,” he wrote.
Anthropic, which opened an office in Seattle in 2024, said it used Claude Code to build Claude Cowork itself. However, analysts at William Blair issued a report this week expressing skepticism that other businesses will simply start building their own software with these new AI tools.
“Vibe coding and AI code generation certainly make it easier to build software, but the technical barriers to coding have not been the drivers of software moats for some time,” they wrote. “For the most successful and scaled software companies, determining what to build next and how it should function within a broader system is fundamentally more important and more challenging than the technical act of building and coding it.”
For now, Claude Code is being rapidly adopted. The tool reached a $1 billion run rate six months after launch in May. OpenAI’s Codex and Google’s Antigravity offer similar capabilities.
“We’re excited to see all the cool things you do with Claude Code,” Caleb John, a Seattle entrepreneur working at Pioneer Square Labs, told the crowd. “It’s really a new era of software development.”
Editor’s note: This story has been updated to reflect that the report cited was from William Blair.
Anthropic's agentic tool Claude Code has been an enormous hit with some software developers and hobbyists, and now the company is bringing that modality to more general office work with a new feature called Cowork.
Built on the same foundations as Claude Code and baked into the macOS Claude desktop app, Cowork allows users to give Claude access to a specific folder on their computer and then give plain language instructions for tasks.
Anthropic gave examples like filling out an expense report from a folder full of receipt photos, writing reports based on a big stack of digital notes, or reorganizing a folder (or cleaning up your desktop) based on a prompt.
AI coding assistants are no longer just autocompleting lines of code, they are quietly making decisions for you. Tools like Claude Code are able to read projects, plan multi-step changes, install dependencies, and modify files with minimal human oversight. To make this possible, these assistants rely on plugin marketplaces, where third-party developers can enable ‘skills’ that teach the agent how to manage infrastructure, testing, and dependencies. Though powerful, the model requires a high degree of trust, thus bringing with it a new set of risks.
At a first glance, third-party marketplace plugins are harmless productivity boosters. Connect a marketplace and enable a plugin so your coding assistant becomes smarter about your stack. However, beneath the convenience is a security blind spot: These same skills often run with extremely high privilege and very little transparency on how they make decisions or where the code and dependencies are coming from. The code issue isn’t prompt manipulation or social engineering – it’s compromised automation.
A full technical blog post by SentinelOne’s own Prompt Security team breaks down how a single benign-looking plugin from an unofficial marketplace exposes a dependency management skill. When the developer asks the agent to install a common Python library, that skill quietly redirects the install to an attacker-controlled source, ensuring a trojanized version of the library is pulled into the project. While nothing looks wrong – the library imports cleanly, the example code runs without error – malicious code is now embedded into the environment, capable of exfiltrating secrets, monitoring traffic, or lying dormant until it is triggered at a later time.
What makes this especially concerning is persistence. Marketplace plugins are not one-off interactions. Once enabled, their skills remain available across sessions and will continue to shape how the agent behaves in the future. Rather than a ‘bad prompt’, this effect is more like compromising your package manager itself.
As AI-driven development workflows accelerate, plugin marketplaces and third-party skills are now part of the software supply chain whether teams realize it or not. If your coding assistant can fetch and execute code on your behalf, every plugin installed joins your trust boundary.
Read the full blog post here for a detailed walkthrough of the attack mechanics and learn why dependency skills are such a powerful, but under-modeled, risk.
Third-Party Trademark Disclaimer:
All third-party product names, logos, and brands mentioned in this publication are the property of their respective owners and are for identification purposes only. Use of these names, logos, and brands does not imply affiliation, endorsement, sponsorship, or association with the third-party.
Login
