Miles Konstantin spent eight years as a creative technical director, conceptualizing ideas and hiring engineers to build them. Now as Harmonic’s Head of Automation & Tooling, he’s the builder. With 15 agents, 100+ workflows, and 33 internal apps (including a very expensive SaaS tool replacement) to his name, he’s helping to give his company capabilities that rival companies exponentially larger.
His path here started in a surprising place: running the internet’s most popular fan blog for musician Childish Gambino.
In high school, Miles was a massive fan of Donald Glover (stage name Childish Gambino). He made a blog about him with some friends, and it became one of the most popular fan sites on the internet. When Glover’s website crashed during an album rollout, Miles saw an opportunity.
“I messaged him saying, ‘Hey, my roommate is a computer science major, we can help you fix this,’” Miles recalls. “So he brought us in. We basically revamped the website, relaunched it, and then my roommates and I became his tech team.”
Throughout college, Miles studied physics by day. By night, he helped to build apps that supported interactive concert experiences, sitting between the engineers who could build and the artist who had the vision, translating between the two worlds.
After college, Miles moved to Los Angeles to work with Glover full-time as his creative technical director. He could conceptualize ambitious ideas and hire engineers to build them, but he couldn’t build them himself.
The barrier was always the same—getting an IDE running, managing dependencies, understanding the entire development environment. “You get so in the weeds that there’s such a barrier to entry,” he says. “It kind of slowed me down.”
Miles joined Harmonic in 2021 as a generalist. The startup discovery platform helps venture capitalists and go-to-market teams find and evaluate startups, but Miles’ role was less defined—he could bridge different worlds but lacked depth in any particular technical area.
The turning point for Miles came with automation tools. First Zapier, where he could string things together visually. Then someone on the team spun up Retool for an experiment, and Miles ran with it.
Harmonic had built its customer management system inside Harmonic’s core application: an admin portal accessible when the team logged into the app. The problem was that every new feature request, every permissions change, and every adjustment pulled engineers away from product development.
Miles decided to rebuild the entire system in Retool. It became his first real project—moving everything over, learning how the internal APIs worked, and figuring out how to build an app inside the Retool interface.
That system is still in use today. It freed up critical engineering resources and proved a crucial principle: the right tool can transform someone from product thinker to builder.
Miles’ growth as a builder coincided with another revolution: large language models. In the beginning, he’d either ask engineering to make changes and wait, or bother someone to teach him how to do something in Python. It was slow.
With LLMs and Retool together, something clicked. Once he understood how an API worked generically, he could do almost anything. The Retool interface made it easy to build applications for the team, and then it was just a matter of fetching data—which Retool also made straightforward. With AI, Miles is able to validate and understand error messages to get unblocked.
But Miles is careful to distinguish his approach from pure “vibe coding”—generating code without understanding it. If you treat it as a complete black box, just having it do something and putting it in without thinking, you won’t learn anything.
“But if you’re iterating over and over and going through similar problem sets or challenges, I think you end up picking up a lot of things,” he says. “The combination of Retool Workflows and LLMs allows you to think at more of a systems architect level, where you’re designing the structure and having it help you with the individual components, but you still have to understand how the data flows through and what’s a good structure for it to follow.”
Miles’ most dramatic demonstration of his new capabilities came when a third-party tool that cost roughly $20,000 annually kept failing. Errors were opaque, and its support function was slow.
Frustrated, Miles made a decision: it would be faster to rebuild its entire product inside Retool Workflows than to wait for their support team to respond. Over the course of a week, while waiting to hear back, that’s exactly what he did.
The original tool integrated with Gong, Harmonic’s call recording software, to process sales calls. It extracted insights, pushed data to their CRM, and automated various tasks.
“In retrospect, it’s crazy that we were paying so much for a tool that did that,” Miles reflects. “But I think at the time it seemed like anything AI must be expensive.”
But Miles recreated all of it—and improved it.
The old vendor system was a black box. When things failed, Miles couldn’t see where or why. With Retool Workflows, everything became transparent. Each step of the workflow is visible, each error traceable—and it was cheaper.
“I think it was on the order of a fraction of a cent per call for us to process through Retool,” Miles says.
The rebuilt system does more than the original. When a Gong call completes, Workflows generate transcripts, create AI-generated summaries for Slack, and route information based on content. Different workflows are triggered for competitor mentions, bug reports, or feature requests.
The system also generates pre-formatted Linear tickets that reps can file with a single emoji reaction, dramatically reducing the manual burden on the sales team.
For Miles, without a traditional engineering background, the visualization is what makes it work. Each step of the workflow is like a concrete block that you can customize. You understand what goes in and what comes out. What would be really complicated thousands of lines of code becomes little blocks you can just see.
“Even someone who would be intimidated by programming can just play in this sandbox and focus on one thing at a time.”
“You can just focus and say, ‘Okay, this code needs to run JavaScript, but I don’t have to think about all the complexities of JavaScript. I just need to figure out what comes in and in what format I want the data to come out,’” Miles explains. “That becomes so tangible that even someone who would be intimidated by programming can just play in this sandbox and focus on one thing at a time.”
When Retool Agents launched, Miles had already built “Hexy”—an internal Slack bot named after Harmonic’s hexagonal logo—using a complex workflow architecture. The bot could answer natural language questions about customers by querying multiple systems and making intelligent decisions about which data to fetch.
Building the workflow version took some time—a week or two to get tested and up and running. With Agents, Miles rebuilt the entire system in about a day.
Beyond the speed improvement, Agents fundamentally changed what was possible.
In the original workflow, Miles had to be prescriptive about everything. He’d specify that the system could do three requests here, then on a second attempt, do another three. Every decision path had to be manually structured, with every scenario anticipated.
With Agents, the system makes its own decisions. It reads through its available tools and chooses based on context. It can decide to check Salesforce to get a customer’s ID, then use that to do another lookup, thinking through the steps on its own.
Now Miles builds specialized “micro agents” that function as tools for Hexy—the main overseer agent that coordinates everything. One example: integrating their competitor comparison documentation from Notion, so Hexy can answer sales questions by pulling from the most current competitive intelligence.
Agents have a finite amount they can hold in their context window at once. Miles’ solution is to have Hexy trigger sub-agents that are very task-specific with specific context and specific capabilities. The sub-agent processes a bunch of data, but then condenses its output before feeding it back to Hexy. This way, Hexy doesn’t need to read through all the logs of some system—it can ask a question, let the sub-agent research and make a determination, then get a concise answer back.
Perhaps most intriguingly, Miles has given Hexy meta-awareness of its own capabilities. The agent can file its own feature requests in Linear when it encounters limitations in acting on something an internal user wants to do. The tickets go directly to Miles, giving him insight into what tools would be most valuable to build next.
What once required vendor evaluations, procurement processes, and engineering time, is now a simple question: Why can’t we just build this in Retool?
When someone at Harmonic wants to buy new software, that’s now the default question. A huge shift came from understanding that what seemed like walled garden systems—separate from each other, requiring special tools to interact with—are really just APIs. Once you realize you probably already have access to those APIs, you can do things quickly on your own. Retool becomes a hub to interact with all these systems in a reliable way and then build on top of them.
The democratization extends beyond Miles. Harmonic’s growth team is able to self-serve some of their needs, and Miles helped a product manager with no technical background build Retool Workflows to interact with tickets, fetch ticket data, run analysis, and send Slack alerts.
“Once you teach someone how the basics work, it’s very cool to see other people without technical backgrounds be able to build things.”
Because Harmonic is still small, Miles takes the opposite approach from traditional software development. Rather than extensive upfront requirements-gathering, he talks to people to understand their pain points, then immediately ships something.
“We’ll say ‘is this working? Let’s refine it rather than doing the back and forth of talking about it in hypotheticals for a long time,” he explains. The low barrier to entry makes this possible.
With traditional software, you’d want a robust process of understanding requirements and limitations. But with Retool, the Harmonic team can have a meeting and then 20 or 30 minutes later, Miles can spin something up and ask: does this solve the problem for you, where are the rough edges, let’s just ship it. There are actually a ton of different systems throughout the company built this way—super rapid iteration where it’s better to ship and get feedback in reality than to hypothetically discuss what people want.
Miles’ journey reflects something larger about how software development is changing. The barriers that once separated those who could build from those who understood the business problems are collapsing.
His transformation from doing one-to-one customer implementations to building scalable tools has allowed Harmonic to punch above its weight.
Miles’ advice for other builders just starting out? “Try Retool. You can play around with so many different pieces of building things—from apps to workflows to agents. Each has a slightly different mechanism, but you’re in one app that, once you get comfortable with it, makes translation easy. Your output becomes so much faster.”
Harmonic is an AI-powered startup discovery platform based in New York City. The company helps thousands of investors worldwide discover and evaluate startups, with additional use cases for go-to-market teams and recruiters.
Reader
