Written by: Techub News Organized
In the latest episode of the YC Root Access program, host Aaron conversed with two co-founders of the startup Conductor, Charlie Holtz and Jackson Sippe. The company has just completed a $22 million Series A funding round co-led by Spark and Matrix. They have developed a revolutionary Mac application aimed at enabling engineers to command multiple AI programming agents to work together simultaneously, thereby greatly enhancing development efficiency. This conversation delves into their entrepreneurial story, product philosophy, and deep insights into the future of software engineering.
From "Building Tools for Themselves" to Changing How Engineers Work
The core product of Conductor is a Mac application that allows developers to run multiple coding agents simultaneously, such as Claude Code or Codex. Unlike traditional methods of manually editing code in an IDE or interacting with a single AI in a terminal, Conductor integrates all coding agents into one interface. Users can simply click a button to create an isolated copy of a code repository, assign AI agents to complete tasks, and then review and merge their work results.
Charlie Holtz and Jackson Sippe met in college; despite their significant age difference, their shared interests and project collaborations brought them together. After graduation, Charlie Holtz was responsible for growth and engineering at Replicate, while Jackson Sippe worked on the machine learning infrastructure team at Netflix. They maintained the habit of collaborating on projects and eventually decided to start a business together.
Their entrepreneurial journey has not been smooth sailing. When they first applied to Y Combinator, their idea was "to help you book restaurants and tennis courts using AI," but this quickly proved to be a case of "searching for a problem to solve." After entering the YC incubator, for the first month and a half, they continually experimented with and discarded various ideas, ranging from B2C to B2B, creating numerous prototypes, but user feedback was often poor. Host Aaron recalled that every few days they would meet and have an entirely new direction.
A turning point occurred during an office meeting. Aaron suggested they consider the development tools sector and gave a crucial piece of advice: "Put a poster of both of you on the wall that says 'Build what these two guys want.'" They actually did it. This advice narrowed the scope to development tools and brought them back to the original intention of addressing their own pain points. At that time, they felt existing AI programming tools were not powerful enough, and the release of models like Sonnet 3.5 presented tremendous untapped potential.
They had tried to build an "IDE after" user interface that would let artificial intelligence take over the workflow completely, but at that time, the model capabilities and tool invocation maturity were insufficient to support this vision. After temporarily setting aside this ambitious idea, they decided to first create a better chat application called Chorus, allowing users to converse with multiple models simultaneously. It was during the development of Chorus and its development tools that they accidentally discovered the prototype of Conductor.
When Charlie Holtz first quickly built a prototype of Conductor and demonstrated it to Jackson Sippe, the latter's reaction was, "First of all, this is not at all what I wanted; a lot needs to change. Secondly, this is so impressive, it works so well, we have to do this." From writing the first line of code to releasing the product took about three weeks. The first week was entirely Charlie Holtz completing the prototype solo, achieving unexpectedly good results; then they spent about two and a half weeks refining it.
The moment that truly convinced them "this is it" was when Charlie Holtz first assigned a task to one agent and then pressed a shortcut key to assign a task to another agent, and then saw the first agent show a "unread" notification point, only to discover it had completed its work. That magical experience of being able to run multiple agents simultaneously and remain efficient, along with Jackson Sippe's feedback—who typically holds a skeptical view on novel ideas—that it was indeed useful, became a key signal. Even more exciting was the fact they could "use Conductor to build Conductor," which in itself created a magical positive feedback loop.
Conductor Cloud: Breaking Local Bottlenecks for Continuous Collaboration
In the interview, Charlie Holtz and Jackson Sippe announced the release of Conductor Cloud. Previously, Conductor operated entirely on the user's Mac locally. This meant that once the laptop was shut down, the working Claude Code or other agents would stop. The launch of Conductor Cloud changed this situation, allowing users to create cloud-supported workspaces where agents could continue functioning even after the computer was turned off.
This solved a major bottleneck users faced: the limits of local computing power and management capability. Many heavy users of Conductor have reached the upper limits of the different working trees or instances they can run locally. Charlie Holtz admitted that, given his personal cognitive load, he could effectively manage at most three to five agents simultaneously. He believes this partially relates to an interface design issue and is a challenge the team is eager to tackle next. They have already proven that running multiple coding agents simultaneously is feasible, but to break through the three to five agents limit and reach a higher level, innovation at an abstract level is needed to reconstruct the interaction interface.
Looking back at the birth of Conductor, they initially tried to manually manage multiple agents, but the friction was too much, making it hard to exceed one or two. Even earlier attempts to build similar tools failed due to immature timing—the IDE could not yet be fully escaped. AI agents needed to reach a "sweet spot" state: where they didn’t require human attention for every action while still necessitating enough human oversight to ensure effective orchestration and multitasking. In early experiments, they even cloned five code repository copies for each agent but later discovered and used Git's worktree feature, gradually building the foundational components needed for Conductor.
The business growth has proven the product's market fit. Since January of this year, the number of Conductor users has grown nearly tenfold. The user base is very diverse, ranging from independent hackers to engineers at large publicly traded companies, who use it daily. While the team is lean, it is rapidly growing and plans to scale up.
Viral Growth and Insights into Top Engineers
In addition to exceptional product-building capabilities, the Conductor team, particularly Charlie Holtz, is also known for their ability to create viral content. One of his notable cases was on the day of the GPT-4 Vision release, where he wrote a simple Python script to observe his actions in real-time through a webcam and narrated it in the voice of David Attenborough, as if he were in a "Planet Earth" documentary. This tweet unexpectedly went viral, even catching the attention of Attenborough himself, although the latter did not particularly appreciate the work.
Charlie Holtz attributes part of this ability to his work experience at Replicate, where his primary task was to drive growth by building demo applications and creating videos showcasing technical possibilities. His insights include: be authentic and direct, avoid corporate jargon, don’t put on airs, and express like you’re chatting with a friend; the content can be simple and straightforward, like his Attenborough video which was a raw screen recording with minimal editing; most importantly, conduct numerous experiments, publish a lot of content, and observe what resonates to grasp the form of "viral content." Jackson Sippe added a succinct conclusion: "Make content that you would also want to click on," which aligns with YC's advice of "build the product you want."
To gain deeper insights into users, the Conductor team often conducts "bike tours" in cities, engaging with users face-to-face to observe how they use the product. This allows them to closely witness how the world’s best engineers build at the technological frontier. They’ve discovered several intriguing phenomena:
- Valuing Skill Documentation: Top engineers invest a significant amount of effort into building and maintaining "skill documentation." They organize best practices for React, specific knowledge about codebases, etc., into Markdown files for AI agents to continuously learn from, thereby enhancing the quality of agent output.
- Configuration Trends Towards Simplicity: Surprisingly, many excellent engineers have very simple and "clean" development environment setups. Charlie Holtz compares this to those with complex Vim configurations who may not necessarily be producing efficiently, implying that over-optimizing tools may not be the key.
- Divide "Free Zone" and "Rigorous Zone": They have observed and are practicing a pattern: in certain parts of the codebase, AI can be allowed to play freely; whereas in other core or design-intensive areas, humans act as architects while AI serves as executors. This creates "clutter-free zones," balancing efficiency with control.
Looking Ahead: From Managing Agents to Leading AI "Legions"
When asked about the future of programming and engineering, Charlie Holtz and Jackson Sippe shared their perspectives. While predicting the future several years out is difficult, they are confident that models will become tenfold or hundredfold smarter than they are now, able to run longer without humans’ frequent intervention, increasingly feeling like a human colleague, even though their "thinking" processes remain distinctly different from humans.
The launch of Conductor Cloud is precisely to embrace this future, allowing agents to run for extended periods. They envision that future interfaces will need to make engineers feel like CEOs managing a massive enterprise with thousands of AI "employees" working for them. Most of the time, you would sit at your desk like a CEO, receiving high-level reports from AI acting as "Engineering Managers," "Marketing Managers," etc. Occasionally, you will need to dive into the details for checks, similar to current code reviews (PR Review), except now you are probing downwards, asking a specific "AI employee" what work it has completed.
This circles back to the previous observation about top engineers: the best AI users are often also the best human collaborators. They can understand what happens at different levels, check agent operations at each layer, optimize bottlenecks encountered, and think proactively instead of passively responding to agent outputs. They need to consider: to complete this task, what design questions do I need to answer? What parts do I need to care about? What constraints should I impose on the agents?
Regarding the future of Conductor, their vision is to make it not limited to personal laptops but capable of running anywhere, becoming the "command center" for managing all AIs within an organization. Users can access the system from their phones, laptops, or existing cloud infrastructure. Charlie Holtz is particularly excited about "raising the level of abstraction." He believes the current Conductor still bears the imprint of its birth period (July 2023) in terms of model capabilities, and he is eager to explore higher-level abstractions that allow people to control models in ways different from now, creating abundant space for interface and product experimentation.
What will this unlock for users? First, it will break through the bottleneck of human working memory (currently three to five agents). Second, it will overcome the bottleneck of code reviews. Currently, engineers still need to very carefully inspect every line of code added by the AI, which consumes a lot of energy. The team believes the current GitHub PR review model is still stuck in the 2010s, and there is massive potential for innovation in the review process in the future.
Finally, when it comes to team building, their current recruitment standards are somewhat "non-traditional": existing team members either play ultimate frisbee or are French. Of course, more importantly, they are looking for those who are agile in action, have startup experience or have founded companies themselves, eager to act quickly and be on the technological frontier. The key is to have the motivation to build tools for themselves and the best engineers. The ability to "use the tools you've built to build this tool," working simultaneously at different levels—both developing specific features and contemplating and improving the underlying workflow to productize it—this experience is truly the core source of joy and motivation for the Conductor team. As Charlie Holtz said, "There is no other work I would rather be doing, and no other people I would rather work with." In this era of dramatic change in software engineering paradigms, working to change the way building itself occurs is undoubtedly one of the most exciting things.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
