How Small Audio Companies Use AI Agents to Survive w/ Wojciech Ozimek (Audio Product Academy) | Ep 42

Three Things You Can Take Away From This Episode

• AI agents can give small audio companies the capacity they could never afford to hire, but only on top of skills they already have.
• Support, updates and compatibility are part of the product, which is why a vibe-coded clone is not the same thing as the software it copies.
• Before you build an agent, give it a clear goal and the right memory, and respect the context window, because most "hallucinations" come from how the tool is used.

Wojciech Ozimek joined The Audio Programmer Podcast from Poland. He spent almost 2 years on Avid's technical program management team shipping Pro Tools, working mostly with the Avid Inner Circle plugin and app ecosystem. Today he runs Audio Product Academy, a course platform and meetup community for the business side of audio, and Augmented Pipelines, which helps audio and entertainment brands put agentic AI to work.

Josh and Wojciech covered a lot of ground: open-source firmware and hardware platforms, what vibe-coded clones leave out, why paying people properly matters, and how Wojciech builds AI agents for product teams. Running through all of it was one question the industry keeps asking. With AI tools everywhere and nobody sure where the market is heading, how do you use AI without losing what makes your work good?

1. Nobody Knows Where the Plugin Market Is Going

Wojciech talks to a lot of founders and product leaders through his podcast and meetups in Berlin, Helsinki and Stockholm. What he hears is contradictory. Some say plugins are becoming a commodity in a saturated market. Others say the market is still growing faster than they can keep up with.

His own position is "safely in the middle." As long as you deliver value, there's room for you. That value might be strong DSP and research, or an interface people love using. He's also candid that consumer products are hard to read, and that many successful launches come down to good timing and some luck. You can't know whether an idea is good until you test it.

2. Support Is Part of the Product

When someone posts a vibe-coded clone of Pro Tools or FL Studio, it can look like the real thing. Wojciech's response draws on a product management idea: every product has 3 layers. There's the core product, the actual product, and the augmented product, meaning the guarantees and support around it.

The augmented layer is where much of the cost sits. Keeping software relatively bug-free and compatible with every macOS update takes real people. Josh shared a Saturday-night story from his own agency: a plugin they'd built for a high-profile client suddenly asked to be registered again in the middle of a session for a very famous artist. When the stakes are that high, people pay for the promise that someone will pick up the phone and fix it.

3. Your Customer Decides How Much Bugginess You Can Get Away With

Some companies ship fast and iterate. Others run an army of QA before every release. Wojciech's point is that neither approach is right in general. It depends on who you're serving.

Bedroom producers and semi-pros will often live with a bug and a workaround. A film mixer on a tight deadline won't, and neither will a studio working with a big-name artist. Before you decide how fast and loose to ship, be clear about which customer you're building for and what you've promised them.

4. Paying People Properly Gets You Their Best Work

Josh explained why he turns down offers of free work, even from people who love the Audio Programmer. If you're not paying someone, anything you ask happens on their time, and there's no exchange of value. If you pay them badly, you can't expect their best work either.

The same logic applies to customers. Wojciech pointed out that ours isn't a wealthy industry, and many music makers don't see the cost of updates, support and salaries behind a product. Some companies do go "to the dark side of the force" on pricing. Charging for upkeep usually pays for keeping the product alive.

5. AI Agents Can Be a Lifeline for Small Audio Companies

This is Wojciech's main argument. Audio is a small market. Product managers call it total addressable market: there are only so many people who mix music or play a particular instrument, no matter how good your product is. That keeps a lot of indie plugin companies small and stretched.

Agentic AI can automate work that small companies could never afford to hire for anyway. Wojciech is clear about the limits. AI-written marketing copy isn't good enough for communication you want people to love, and nothing about it is a fully autonomous business. But for repetitive business processes, it gives a 2-person company capacity it didn't have before. His own platform is run by him and his wife.

6. AI Won't Replace a Skill You Don't Have

Wojciech started experimenting with AI to prove it was rubbish. He ended up teaching it. The lesson he keeps coming back to is that you can't be a great AI product manager without being a great product manager first. You can't automate analytics if you don't know which data matters.

Josh made the same point from the DJ world. When DJs moved from vinyl to digital systems, some got bitter and others became the pioneers of the next generation. The people who already understand what's happening under the hood, and are willing to adopt the new workflows, end up with something close to superpowers. Josh's friend Jaguar Skills makes an AI-generated ninja series, Full Metal Ninja, that works because he knows cinematography and culture deeply. The tool doesn't supply the taste.

7. Friction and Teamwork Are Part of Making Something Good

Both of them pushed back on the promise of a frictionless creative process. Learning an instrument or a martial art takes effort, and people value the story behind a track or a product. Wojciech cited Elektron, whose instruments ended up with workflows that became their own creative language.

Josh admitted that as a developer he could prompt his way to a plugin built exactly his way. He doesn't, because his team mixes outside-the-lines thinkers with people grounded in music tech tradition, and that mix produces something neither side would make alone.

8. Watch Out for AI Silos

Wojciech sees a new problem inside teams. People work through an idea with their own AI, arrive at the meeting with their mind made up, and the brainstorm never happens. He calls these AI silos.

His answer is to make agents part of the shared process. Put them in Slack or Teams channels, give them a common memory such as a decision log, and stop routing everything through one person's private chat. Josh described a similar setup: his "Josh OS" keeps standing decisions for each of his ventures, so the rules sit right next to the work.

9. Build Your First Agent Around a Goal, Not a Tool

Wojciech's course, Building Your First Agent for Product Work, is aimed at product teams and founders more than engineers. It starts with fundamentals, and the first one is a goal. A team that asks for "an agent for Jira and Confluence" has named its tools. A team that says "take a client's PRD, break it into user stories, and estimate them" has told the agent what it's for.

From there you build the master prompt, then memory files about yourself, your company and your methodology, and only then add connectors and skills. He also stresses the context window. Load too much into it and the model starts compacting and making things up. Wojciech calls that a usage problem. His rule of thumb is to talk to AI the way you'd brief a new employee: if a person couldn't follow your instructions, neither will the model. And treat every agent as a small product that needs maintenance as models and API keys change.

Closing

The conversation never pretended to have a neat answer. Josh's view is that how any of us feels about AI matters less than how we adapt, and Wojciech agrees, as long as the fundamentals come first. His last piece of advice had nothing to do with AI: get up from the computer, go to the studio, and watch how music makers actually work. The tools keep changing. Understanding the people you build for is still the job.