Claude Engineering Lead: Why Giving AI Full Access Changes Everything

Summary

Felix Rieseberg, who leads engineering work on Claude products at Anthropic, argues that many people still use AI for narrow tasks—such as summarizing email—when it can take on broader goals. His central recommendation is to give Claude the context, tools, and access needed to complete a workflow, then judge the result rather than micromanaging every step.

How Felix Works With AI

• Felix now spends much of his time prototyping: testing ideas quickly, comparing possible designs, and using Claude to help build prototypes that would previously have required more time or a larger team.
• AI changes the cost of trying an idea. Instead of first debating whether something is feasible or worth the engineering effort, teams can often build a prototype and evaluate it.
• He distinguishes automation—doing an existing task faster or more cheaply—from augmentation—enabling people to do things that previously weren’t practical. He sees the latter as especially compelling.
• AI can make building easier, but it does not remove the need to decide what is worth building. A successful product still needs focus, usability, and a real understanding of users’ needs.

Getting Better Results From Claude

• Give it the goal, not just a small instruction. Felix describes moving from manually selecting a crash report for Claude to investigate toward giving it access to the relevant reports and asking it to identify and resolve issues.
• Provide useful context and tools. Instead of loading every piece of data into each prompt, he favors making appropriate information available through tools and workflows. Limited access can prevent Claude from understanding the broader business context.
• Start with the standard tools. His suggested first step is to try Claude’s desktop app and its computer or browser capabilities, then add connectors or other integrations when a workflow calls for them.
• Don’t overcomplicate agents. He treats agents less as special personas and more as Claude working on different kinds of tasks through Projects and other workflows. Simply telling a model to “act like an expert” is not a substitute for giving it the relevant tools and information.
• Use skills selectively. Skills can help, but guidance for older models may not suit newer ones. Sometimes a straightforward prompt works better, so Felix advises evaluating whether a skill actually improves the result.

Reviewing AI Work and Catching Mistakes

• Confident-sounding output should not be treated as proof. For factual work, Felix recommends checking primary sources and judging the accuracy and usefulness of the evidence—not just whether the answer sounds plausible.
• Adversarial review can help: ask another model or session to find flaws, challenge assumptions, check facts, or make the strongest counterargument. This is useful for drafts, documents, slides, and code.
• Different models may give conflicting feedback, so a human still needs to decide which critiques are valid and actionable.
• For creative work, “good” is harder to measure than for tasks with clear factual answers. Felix notes that skilled writers and artists often follow deliberate, iterative processes. AI can assist with those processes, but evaluating taste and quality remains challenging.
• He recommends a side-by-side trial when trust is uncertain: do the work yourself while Claude does it, compare the results, and use the differences to improve the workflow.
• In his own work, outputs still pass through human review. He says he does not fully trust an agent to verify every fact without further checks.

AI at Work and in Product Development

• Felix describes his role as a blend of product, engineering, people management, and data work, with more attention now going to prototyping and human considerations.
• His team uses Claude to prepare podcast research, including reviewing previous interviews and identifying questions or topics that may be overlooked. The host says Dig AI was added to their workflow to surface patterns in public videos, comments, and audience discussions.
• He sees an opportunity to improve outdated business software. Many organizations still rely on clunky systems, and he believes more capable AI applications could address those problems beyond basic FAQ-style chatbots.
• As software becomes easier to produce, he believes execution is becoming cheaper. Understanding customer needs, showing empathy, and identifying valuable problems remain important advantages.

Advice for Engineers and Founders

• Felix says junior engineers can be especially willing to experiment with AI and use it to attempt more projects. He sees opportunities for small teams to build and test ideas quickly.
• For engineers, the measure of good work is its impact on users—not merely technical elegance or mastery of a particular framework.
• His career advice is to build a real portfolio project and demonstrate both the ability to make something and the ability to understand what customers need.
• He suggests that combining systems and tools effectively is becoming more valuable as AI makes implementation easier.

Guides and Practical Recommendations

• First AI agent guide: The host says she prepared a guide for setting up a first agent in about 20 minutes, available through her newsletter.
• Getting started with Claude: Felix recommends beginning with Claude’s desktop app and its built-in browser and computer capabilities, then expanding access or adding connectors as useful.
• AI review workflow: Collect feedback, ask Claude to summarize recurring principles, and use those principles as input for future reviews. Keep human judgment in the loop, especially for consequential work.

Main Speakers and Sources

• Felix Rieseberg — Anthropic engineering lead; discusses Claude, AI-assisted engineering, product development, and how to evaluate AI output.
• Silicon Valley Girl host — Interviewer; describes the show’s AI-assisted podcast research workflow and mentions Dig AI.