As a technologist deeply embedded in the world of artificial intelligence, Laurent Giraid has a unique vantage point on the seismic shifts reshaping the software industry. His work focuses on the practical application of machine learning and natural language processing within the enterprise, giving him a front-row seat to the AI revolution. Today, he joins us to discuss the incredible momentum behind Anthropic’s Claude Code, exploring the technology that allows a simple Slack message to become a functional piece of code. We’ll delve into the staggering business growth fueled by this innovation, the profound, and sometimes challenging, impact it has on an engineer’s daily life, and the ambitious vision for a future where writing software is as natural as having a conversation.
It sounds almost like magic—a product manager reports a bug in Slack, and a little while later, a pull request appears. Could you pull back the curtain for us and describe the journey Claude takes, from the moment it’s tagged in a conversation to delivering that final link?
It really does feel like magic, but the mechanics behind it are a thoughtful orchestration designed to eliminate friction. When someone types @Claude in a Slack thread, the first thing that happens is an analysis of the intent. The system is surprisingly adept at discerning whether the conversation—a bug report, a feature idea—is an actionable coding task. Once it makes that determination, it kicks off a dedicated Claude Code session behind the scenes. It doesn’t just start with a blank slate; it intelligently pulls in the recent messages from that Slack channel or thread to build a rich contextual understanding of the problem. From there, based on repositories you’ve already authenticated, it figures out which codebase to work on. As it investigates and writes the fix, it doesn’t just go silent. It actually posts status updates right back into the original Slack thread, so the team sees it making progress. The final step is delivering that wonderful little package: a link to the complete session for review, and a direct link to open the pull request. It closes the loop entirely, keeping the solution right next to the initial problem.
Hitting a billion dollars in annualized revenue in just six months is staggering. What deep-seated pain points in the enterprise world do you think are driving this incredible demand, and how does bringing a tool like the JavaScript runtime Bun in-house fit into your strategy for staying ahead?
That kind of velocity speaks to a fundamental and unmet need in enterprise software development. Companies like Netflix, Spotify, and Salesforce are in a constant race to innovate, and the gap between discussing a problem and implementing a solution has always been a major bottleneck. Claude Code’s ability to live inside the workflow, right where the conversations happen, directly addresses this. It’s not just another tool engineers have to switch to; it’s an ambient presence. The demand is fueled by tangible results, like we’ve seen with Rakuten, where they slashed development timelines from 24 days down to just 5. That’s a 79% reduction. The acquisition of Bun is a brilliant strategic move to support this growth. Bun isn’t just a component; it’s an all-in-one toolkit that’s dramatically faster than its competitors. By acquiring it, Anthropic isn’t just buying a piece of technology; they’re investing in the core infrastructure that powers Claude Code. It ensures they can scale the speed and performance that their enterprise customers now depend on, solidifying their position not as a feature, but as a foundational platform for modern development.
Your internal research points to a massive 50% productivity jump, but what really caught my eye was that 27% of the work done with Claude wouldn’t have happened at all. Can you give us a sense of what these “nice-to-have” projects look like and how this shift is redefining what an engineer can accomplish in a single day?
That 27% figure is, to me, the most revolutionary part of this whole story. It represents a fundamental expansion of what’s possible. We’re not just talking about engineers completing their assigned tickets faster. We’re talking about the creative and exploratory work that always gets pushed to the bottom of the backlog because it’s not seen as cost-effective. For example, an engineer might now spin up an interactive data dashboard for the marketing team in an afternoon, a task that previously would have required weeks of planning and justification. Or they might engage in exploratory R&D, prototyping an idea that was just a “what if” in a meeting yesterday. It transforms the nature of the work. The productivity boost means an engineer is no longer just churning through a predefined list of tasks. They have the bandwidth to be proactive, to build helpful internal tools, to scale up projects that seemed too ambitious, and to truly innovate in ways that were previously impractical. The daily workflow becomes less about heads-down execution and more about creative problem-solving.
Even with all this power, your own data shows that engineers are hesitant to ‘fully delegate’ more than 20% of their work, and some even miss the old way of collaborating. How do you balance the drive for automation with the critical need for human oversight and address the very real cultural anxieties that arise when an AI becomes the first person you ask for help?
This is the human center of the transformation, and it’s a delicate balance. The data is clear: our engineers report they can “fully delegate” only a small fraction of their work, between 0-20%. This tells us that Claude isn’t a replacement; it’s a collaborator. It requires active supervision and validation, especially for high-stakes code. The cultural shift is very real. We have engineers who say their dependence on their human team has dropped by 80%, but that last 20% is absolutely crucial. On one hand, some appreciate the reduced “social friction.” On the other, you hear things like, “I like working with people and it is sad that I ‘need’ them less now.” There’s a genuine concern about the atrophy of skills—when producing code is so fast, it can be harder to take the time to deeply learn the underlying principles. The best practice we’re developing is to frame Claude as an accelerant for collaboration, not a substitute. It’s the “first stop” for a question, which can then be refined and brought to a human colleague for that crucial final bit of insight, ensuring quality and preserving the essential human connection in teamwork.
You’ve painted a picture of ‘conversational coding’ as the future. Could you break down how a technology like the Model Context Protocol (MCP) is the key to unlocking this, and what the next tangible steps are to move from where we are today to a world where building software truly feels like a conversation?
“Conversational coding” is absolutely the north star. The vision is to completely dissolve the walls between discussing a problem and solving it. Today, the AI ecosystem can be fragmented; connecting an agent to a new tool or data source often requires building a custom, one-off integration, which is slow and inefficient. The Model Context Protocol, or MCP, is designed to shatter that barrier. It’s an open standard, a universal language that allows AI agents and external systems to communicate seamlessly. Instead of building countless individual bridges, a developer implements MCP once in their agent, and it immediately unlocks an entire ecosystem of compatible tools and integrations. It’s the “plug-and-play” standard we’ve been missing. The next steps are about building on this foundation. We’re in a research preview phase with the Slack integration, gathering feedback to refine it. The goal is to make these integrations so robust and intuitive that the developer doesn’t even think about them. The future is one where the AI is an ambient partner, participating in the natural flow of conversation in the chat windows where developers already spend their day, turning words into working code without ever needing to switch context.
What is your forecast for the future of software development in the next five years, as these conversational AI agents become even more deeply integrated into our daily workflows?
Over the next five years, I forecast a radical compression of the development lifecycle, from initial idea to deployed product. The lines will blur completely. We’ll see fewer rigid, siloed roles and more fluid, empowered creators who can leverage AI to handle the rote implementation details, freeing them up for higher-level architectural and product thinking. The productivity gains we see now, like Rakuten’s 79% reduction in development time, will become the industry standard, not the exception. However, I think the most important change will be in the skills we value. Deep-seated expertise will always be crucial for oversight and complex problem-solving, but the most valuable skill will be the ability to adapt. As one of our own engineers wisely put it, “Nobody knows what’s going to happen… the important thing is to just be really adaptable.” That’s the core of it. The tools will change at a breathtaking pace, but the engineers who thrive will be the ones who embrace this new collaborative paradigm with curiosity and a willingness to constantly learn and evolve alongside their AI partners.
