Rethinking AI’s Role in Software Development
As AI tools become more capable at generating code, engineers are being pushed to rethink how they work. Marcel Weekes, Vice President of Software Engineering at Figma, and Jake Albaugh, Developer Advocate at Figma, discussed this shift at the RenderATL conference, outlining a set of skills that go beyond using AI as a simple automation cheat code.
From Cost Cutting to Creative Exploration
Weekes argues that framing AI primarily as a cost-saving measure misses the point. Quoting Box CEO Aaron Levie, he says, “AI is a way to maximize the upside of what you can do—to be more productive and leverage the skillset you already have.” The real value lies in redirecting effort toward the higher-level responsibilities of development: identifying the right problems, building user empathy, and maintaining craft. Albaugh echoes this, noting that automating the tedious parts of the job frees engineers to focus on the “meaningful thing.” He adds that employers are looking for people who understand the value of their work and can collaborate to evolve ideas.
This mindset reframes a practice like vibe coding. Weekes sees its true purpose not as a shortcut for sloppy output, but as a way to “explore the problem space and come up with a better solution.” By using AI to rapidly generate and visualize multiple paths through a problem, teams can parallelize their exploration and get tangible artifacts for feedback. "Instead of considering just one or two paths through the problem maze, people are able to explore more paths in parallel and get visual artifacts that they can then share with someone to get feedback," he explains. Albaugh points out that this frontloading of the design process puts the user experience front and center, ensuring teams are designing the right thing rather than having the thinking automated away.
Providing Context for Smarter Agents
One of the key technical enablers for better AI output is the Model Context Protocol (MCP), a standard that allows AI tools like Cursor or Copilot to talk to other software. Figma’s own MCP server gives LLMs detailed context about design files, which Albaugh says leads to higher-fidelity implementations. “Developers are implementing things the right way, and users are benefiting from anything you’ve baked into your component library on an accessibility front,” he says. The server makes sure that the specific, institutional knowledge of a large component library is communicated to an AI that might otherwise be unaware of it.
LLMs also serve as an effective internal review tool. Weekes notes that Figma engineers sometimes pre-review their own Pull Requests by feeding them to an LLM before submitting them for human review. “Because these LLMs are aware of our codebase, they can find things like: ‘You rewrote something here that already has an implementation here,’” he says. This catches redundant code early and helps boost throughput.
Orchestrating and Experimenting with AI
As AI agents become more prolific, a key engineering skill will be delegation and orchestration. Weekes suggests that the strongest developers are already learning to break problems into smaller parts and assigning them to multiple AI agents, then reconciling the results. This process involves writing detailed markdown files and instructions that give the LLMs the same kind of context you might give a new intern. He expects this to be a defining skill going forward, even if "no one's an expert at this yet."
Ultimately, the most direct path to proficiency is hands-on experimentation. “Use AI tools as much as you can,” advises Weekes. “Play around with them and push the limits, and see where you surprise yourself." Albaugh’s experience with Figma’s MCP server confirmed this; he uses it to get a high-level summary of a design file’s intent, much like he would by zooming out to see the big picture before diving into code. This kind of familiarity is becoming a marketable asset. Weekes points out that a large portion of the software development workforce has yet to leverage AI, so taking the time to learn its capabilities is a distinct advantage in the job market. The successful engineer of this era will be the one who uses AI not just to work faster, but to work smarter on more meaningful problems.
