Versatility at Scale with Carmen Huidobro

🎧 New Señors @ Scale Episode

This week, I spoke with Carmen Huidobro, CTO at Incredible Bee in Vienna, where she builds products and helps teams figure out what should be automated and what should stay in the hands of users. Carmen has spent 17 years in tech, nearly all of it freelancing — a route she took partly because Austrian employment law made it the practical option for someone who moved from Chile. That freelance path took her across Objective-C, the web, mobile, Ruby on Rails, hardware, and a stint writing ABAP at an SAP consultancy, with five years in developer relations along the way.

In this episode, we dig into why versatility beats both generalism and specialization, how she shipped a local LLM inside a twenty-year-old macOS app without compromising the people who use it, what she pushes back on when investors demand AI in the product, the 2016 refugee hackathon project the City of Vienna is still running, and why she won't charge junior developers for a bootcamp.

⚙️ Main Takeaways

1. The Jack of all trades quote has a second half, and it changes everything

Her framing for what actually survives a shifting market.

• The full version: "Jack of all trades, master of none, but oftentimes better than master of one." She finds that genuinely empowering — "stuff comes and goes, best practices come and go, but the fundamentals they stay."
• Not generalism: She's careful that this isn't the generalist-versus-specialist debate. "It's more important to make yourself an expert in adapting. Not so much generalism, but adapting."
• What that looks like in practice: Being open to new ideas, recognizing patterns, and "being able to throw myself into an environment with an understanding of how to navigate a solution while minimizing the chaos that I bring."
• On stacks: "I do strongly believe that there's the right tool for the job. I don't believe that one size fits all, and if it did, it would have been invented by now."

2. Niches never disappear — ABAP is the proof

The best story in the episode, and the point it makes is serious.

• What SAP is: "Software that is older than you and I combined... essentially software for helping businesses do their transactions in a way that replaces pen and paper."
• The language: ABAP, a proprietary language she describes as a mixture of SQL and Java, and means that neutrally. "It is what it is."
• The detail she can't stop laughing about: True is the string X. False is a whitespace. It sounds awful until you see where it comes from — "it's based on pen and paper. It's an unticked box. So in that context, it makes sense."
• The punchline that made the point: She tried using AI while being onboarded onto ABAP. "Poor AI, given the lack of data, I think it understood ABAP worse than I did."
• The conclusion: "Niches will always exist and expertise will always be needed." What you make portable is the pattern recognition, not the syntax.
• Where the portability shows: Objective-C taught her memory management "like nothing else," even with a syntax gnarly enough that a whole website existed purely to copy-paste block syntax from.

3. Ethics first is a rule you set before the feature exists

How she handled a client asking to add AI, without either capitulating or refusing.

• Her honest starting position: "I'll be the first to admit that I haven't hopped on the AI hype train yet."
• What she said when asked: She had an idea where a prompt interface would genuinely be useful — but with rules attached, stated up front. "Number one rule, ethics first. We'll use a model that is as ethical as can be in this day and age. It will run on the client's computer."
• The test she applies: Cogent with her values and providing value to the end user, worked out in a back-and-forth with actual users.
• On investor pressure: She's careful and sympathetic rather than dismissive. "I can sympathize, because investors want that. In order to raise your series XYZ, the investors want that." But the options include rejecting it and finding an alternate strategy — "it is possible to survive without that investment, believe me."

4. Shipping a local LLM into a twenty-year-old Mac app

The concrete build behind the principles.

• The product: Renamer, a batch file renaming tool for macOS that's been around since before 2004 — before she entered tech. Drag in files, apply heuristics, regex, find and replace.
• The real user problem: A lot of the customers aren't developers. "A lot of our support requests are like, how do I do this, can you explain regex to me?"
• The feature: Ask in plain language — rename these into JSON structure, translate these file names into Russian — and let the model do the work.
• The model: Mistral, chosen partly because she's based in the EU. "The Mistral model worked perfectly. When they bring up something better, I've set it up so that I can download a different model and test accordingly."
• The runtime: A llama.cpp runtime integrated into an Electron app, inside the famously restrictive Mac App Store — and later onto Windows, where getting it running across different GPUs was, in her words, "solving a problem in a way that goes beyond the web. It was fun."
• Why the model isn't bundled: An eight gigabyte binary on top of Electron's already-large footprint. It downloads on demand instead.

5. Opt in, not out — and never take features away

The philosophy underneath the implementation, formed by watching a decade of products get worse.

• The rule: "This runs on the client's computer at their own discretion," and it's opt-in by design.
• Where it comes from: "I find myself burned by the model of the 2010s, when free services like Dropbox or Evernote would give you all of this free storage. And over the years, as they've needed infinite growth, they've taken away features. I sort of made it my philosophy based on that to never do that."
• The alternative: "If you need infinite money, add more value, not less."
• The naming argument she lost, and was right to lose: Her instinct was to avoid the word AI and call it "smart renaming." She was given the feedback to be upfront instead. "Let's acknowledge the fact that it's an LLM."
• What the messaging actually says: We don't store anything, this is privacy first, this is all on your computer, you don't want to use it — don't. Plus a link to the model's Hugging Face page "so you can see what the model's all about and make your own informed decision."
• The principle behind all of it: "I've always found it more important to give the user the keys to the power, and to make that as approachable for them as possible without obfuscating."

6. Where2Help: a hackathon project the City of Vienna still runs

The work she's proudest of, and it was unpaid.

• The context: 2016, the refugee crisis, with large numbers of people arriving at Vienna's Westbahnhof. A friend organized a refugee hackathon.
• The problem someone brought: Too many volunteers showing up at once. Coordinators would have to turn people away — "sorry, we don't need any doctors today, can you please come back tomorrow?" — and "those people, feeling deflated, don't come back tomorrow or ever."
• The build: A Ruby on Rails app where NGOs list what they need and volunteers sign up against it. Built at the hackathon, then continued every Wednesday at each other's houses for months afterward.
• What she got from it: "That's where I learned all of my collaboration skills. That teamwork stuff really came together for me during that project."
• The outcome: They open sourced it, went through the bureaucracy, and handed it to the City of Vienna. "That was 2016. It's being used to this day. I haven't touched a line of code of it in 10 years."
• How it evolved: The purpose shifted over time — less refugee assistance, more things like teaching classes — and it kept running.
• Why it beats paid work on her list: "Yes, it was unpaid, but it doesn't matter. I got these connections that I work with to this day, but also it made a difference."

7. Pointless is not useless

On weird software, and why it's not a phase the industry outgrew.

• My side of it: AI has made it possible to be weird again — the internet feeling a bit like the era when everyone built strange little sites, because nothing stops you from building in a stack you don't know.
• Her correction: "I've honestly felt like that never really went away. It just kind of moved." For her it moved into hardware — Raspberry Pi, Arduino, microcontrollers through the 2010s, and lately e-ink displays wired to a microcontroller for photos or a calendar.
• The quote she borrows: From Charlie Gerard, roughly — pointless is not useless. "Just because you made something weird that might not be for everybody doesn't mean it's something bad."
• Why it matters beyond the hobby: "You do learn from it. You do find yourself able to integrate new ideas and bring them into other projects."
• Her balanced verdict on AI: As much as she resists the overhyping for ethical and political reasons, "it's not one without its merits. Once we find that use case that empowers us in a way that doesn't destroy the planet, that's a good thing."

8. Freelance work came from showing up, not from selling

How she actually found clients over 17 years, including the part that's hard right now.

• The start: One main client — Philip at Incredible Bee, an old friend and mentor, whose company she's now CTO of. But working alone at home got lonely.
• What worked: The meetup scene, despite being a shy person. Talking to people, giving presentations, then conferences and speaking. "By being out there and expressing to people what it is that I do in a very clear as day way," people started arriving with projects.
• The collaboration pattern: Hackathons and joining other freelancers on client work, which she notes isn't far off how a lot of contract work happens now.
• The honest present tense: "Full disclosure, now I'm trying to do that again. I don't know if you've seen the economy, not fantastic. I got a mortgage back when DevRel was a lot more lucrative than it is now."
• The consulting she's shaping: Helping teams decide what should be automated and what shouldn't, informed by five years in developer relations and developer marketing — "what do people actually want, and how do they want that communicated to them in a way where we are additive, not a compromise."

9. Stop telling women they have imposter syndrome

The most pointed section, and it ends somewhere unexpected.

• The article she cites: Stop Telling Women They Have Imposter Syndrome, from Harvard Business Review.
• The argument: Intent versus impact. The framing has good intentions but "has actually damaged us a lot, because it has encouraged us to stifle ourselves."
• The asymmetry she's noticed: "You don't get that from folks who are not marginalized or who are less marginalized."
• Her own path around it: Never having had a corporate junior-woman-in-tech experience, because she was always freelance.
• The superpower she built instead: "I can put up a proof of concept and just feed it to the lion, so to speak. And they'll tear it apart." In a room stuck bike-shedding, putting a bad solution on the table unblocks everyone.
• What she calls it: "This is just me being cheeky — I call this white guy driven development. Because you're navigating the world with that confidence. And I don't mean that in a bad way. If anything, it's something that those of us who are less represented in tech should harness."

10. AI won't teach you how to get unstuck

Her case for why bootcamps and mentorship still matter, built on a story she clearly didn't enjoy living.

• What happened: Teaching the freeCodeCamp JavaScript curriculum on a live stream, working through an exercise on a deeply nested object, and getting stuck. "I'm talking like stuck for 20 minutes." Live, with a chat full of people saying they're lost and suggesting changes without specifying a line.
• The aftermath: She solved it, finished the stream, and felt deflated. "I'll admit I cried."
• The message that reframed it: A learner wrote to thank her — for showing that someone who'd been at this for fifteen years can get stuck, and for showing her problem-solving process.
• The lesson she draws: "AI ain't gonna teach you how to solve a problem. AI ain't gonna teach you not to drown in a Python error log." Nor how to handle a customer who sends you a photo of their screen saying they can't register their license key.
• Why she won't charge: "I don't feel comfortable taking money from emerging developers, especially as the junior developer market is quite frankly in shambles right now." She and her friend Jess still run free web development bootcamps at Bad Website Club.
• What she teaches beyond code: The corporate savvy to navigate a hiring landscape that has genuinely changed.
• The payoff: "There is nothing quite like somebody you taught 10 years ago messaging you and being like, I could not have made it to being a staff engineer at Microsoft without you. When you lift others up, you get lifted with them."

🧠 What I Learned

• The full quote is "Jack of all trades, master of none, but oftentimes better than master of one" — and the durable skill is being an expert at adapting, not at breadth.
• Niches like ABAP never disappear, and they're exactly where AI is weakest, because the training data isn't there.
• ABAP represents true as the string X and false as whitespace, because it was modeled on ticked and unticked boxes on paper.
• Setting ethical rules before the feature exists — runs locally, opt-in, most ethical model available — is what makes shipping AI compatible with your values.
• A twenty-year-old macOS app can carry a local Mistral model through a llama.cpp runtime in Electron, downloaded on demand rather than bundled into an eight gigabyte binary.
• Being upfront that a feature is an LLM beats softening it to "smart renaming" — users deserve the model's Hugging Face page and a real choice.
• Never take features away. If you need more revenue, add more value.
• A hackathon project handed to the City of Vienna in 2016 is still running a decade later, untouched, having quietly changed purpose along the way.
• Weird, pointless projects are how you learn patterns you later reuse — and for Carmen that expression moved into hardware and e-ink rather than disappearing.
• Freelance clients came from meetups, talks and hackathons rather than from selling, though the current market makes that harder.
• The imposter syndrome framing disproportionately stifles marginalized people; putting an imperfect proof of concept in front of a room is a way to reclaim that confidence.
• Free bootcamps still matter because AI won't teach you how to get unstuck, how to read an error log, or how to handle a confused customer.

💬 Favorite Quotes

"Jack of all trades, master of none, but oftentimes better than master of one."

"It's more important to make yourself an expert in adapting. Not so much generalism, but adapting."

"Poor AI, given the lack of data, I think it understood ABAP worse than I did."

"Number one rule, ethics first."

"If you need infinite money, add more value, not less."

"I've always found it more important to give the user the keys to the power... without obfuscating."

"That was 2016. It's being used to this day. I haven't touched a line of code of it in 10 years."

"Pointless is not useless."

"I call this white guy driven development. Because you're navigating the world with that confidence."

"AI ain't gonna teach you how to solve a problem. AI ain't gonna teach you not to drown in a Python error log."

"When you lift others up, you get lifted with them."

🎯 Also in this Episode

• Learning to program by making Mortal Kombat on a TI-83 graphing calculator, then HTML and CSS through Myspace and Neopets
• Wanting to be a theater kid, and engineering being the compromise her father would accept
• Sending a cold email about a macOS X job with zero experience, on the strength of having downloaded Xcode — and getting taken under their wing
• Why freelancing made more practical sense than employment in Vienna as a non-EU citizen
• The Ruby on Rails lunch-ordering app she still runs for companies in Vienna's Industriegebiet
• Market Data Cloud, the bootstrapped stocks API she co-founded around twelve years ago with a remote team, back when that was still unusual — and the decision to wind it down
• Her next move: consumer and power user apps built to her own values, as the sole decision maker, with ADHD and decision-making named openly as part of the calculation
• Volunteering at Distribute Aid, which open sources humanitarian aid processes, and her reservations about "do open source, it'll change your life" advice that ignores where you live
• Why she now volunteers her tech skills with a queer organization in Vienna that has nothing to do with her day job

Resources

More from Carmen and the tools mentioned:

• carmenh.dev — Contact, consulting services, blog posts and talk recordings
• Carmen on X
• Carmen on GitHub
• Carmen on Mastodon
• Distribute Aid — Open sourcing processes for humanitarian aid, always looking for volunteers
• Renamer by Incredible Bee — The macOS app with the opt-in local model
• Mistral and llama.cpp — The model and runtime behind that feature
• Bad Website Club — The free web development bootcamps she runs with Jess

Articles mentioned:

• Stop Telling Women They Have Imposter Syndrome — Harvard Business Review

🎧 Listen Now

🎧 Spotify
📺 YouTube
🍏 Apple Podcasts

Episode Length: 42 minutes on versatility as the durable skill, seventeen years of freelancing across wildly different stacks, shipping ethical AI into a legacy Mac app, and why free bootcamps still teach something a chatbot can't.

Whether you're deciding what to automate, being pushed to bolt AI onto a product that doesn't need it, or wondering which of your skills actually transfer, this one has a clear answer and a lot of hard-won evidence behind it.

Happy adapting,
Dan