TL;DR: What I learned in 2024

1. Design your environment to make it easy to do the right things.
2. Find your reps. Piano has scales, you play through the notes. You practice the finger motions. You get used to groupings of notes. You build a mental map of the keyboard. You can work on certain stylistic elements. Whatever field your in, what are the reps you can do to get better? For me in coding, it’s not writing code, it’s debugging and cleaning up code. Every day I try to find at least one issue or a part of the code that I can refactor.
3. Why can’t I do this faster? What is the fastest way to get a working version? What is the easiest way to make significant progress? I try to go to a working version within 3 hours. Then I improve. Make it run. Make it right. Make it fast.
4. The main differentiator is execution. Most people are lazy.
5. What is the biggest lever for what you want to accomplish today? Working hard is not equal to big rewards. Find leverage.
6. Figure out how to differentiate yourself. You want to stick out like a sore thumb. That goes for business and yourself. You want to make it easy for people to remember you and bring you up in a conversation. “The X guy”.
7. Where are you now? What is the next step? What is the end vision?
8. The little tasks add up. Hidden operational overhead can be a major time suck.
9. Niche down. You can’t be everything to everyone.
10. Stay close to existing solutions when building something new. Don’t try to reinvent everything at once.
11. Batch similar work together. When you can focus on one type of task for hours, you learn and improve faster.
12. Make it easy to be consistent.

I had a lot of fun this year. I learned a lot, I built a lot, and I met a lot of interesting people.

I got a lot done in 2024. I achieved most of my goals, but I am not where I want to be.

For people who don’t know me, I run an AI agency called Aisbach with a venture studio (we build startups). On the side, I also have a podcast called How AI Is Built.

In my agency, I continue to shift my focus on building enterprise AI systems. It shifted a lot from LLMs to data engineering and MLOps.

My podcast went from 0 to 40 episodes, from 0 to 750 listens per episode, and from 0 to 350 subscribers (approximately).

I will first review the goals I set for 2024, then set the goals for 2025, and last I will review what I learned in 2024.

Goals of 2024

• [Completed] Publish 30 episodes of the podcast
• [Completed] Make a living from my own companies
• [Failed] Take a longer vacation with my family and girlfriend I went on a few short trips in 2024, but did not make it for longer than 3 days.
• [Completed] Read 30 books Not much to say here. I hit 32 yesterday. But in a lot of them I did not go to the level of depth I would have liked. I will change this goal for 2025. Go for depth over breadth. I read really broadly in 2024 to get a good base in a bunch of different areas of AI and business. Now I think I have a good overview and can focus on the details.
• [Failed] Find a consistent routine for working out I wanted to get back to my old fitness routines. I did not manage to do this. I had a few injuries, illnesses, and a lot of little things, which held me back. I did not gain weight, lose muscle, or gain fat, I mostly maintained, but I am a long shot from the old fitness level of 2023.
• [Completed] Launch a first successful venture In my agency, we started a venture studio. We are still in the early stages, but we successfully launched our first venture. We scaled it to a few thousands in MRR, got an acquisition offer, and got a lot of customer feedback to make it better.
• [Failed] Write 10 technical blog posts with accompanying code I did not manage to write a single post. With all the coding in the agency and the venture studio, I did not have the time to write much. This is on the agenda for 2025.
• [On Track] Network with 10 of my idols I met a lot of interesting people, especially through the podcast. I got to meet 4 of the 10 people I wanted to meet in the next 5 years (2024-2029). I might have to make it more ambitious for 2025.

What’s coming up in 2025

I want to make my goals a little bit more specific and ambitious.

• [New] Build in public

  I want to build in public more. I want to share more of my learnings, my process, and my thoughts.

  The concrete goals are integrated into the other content goals.

• [Renewed] Publish 52 episodes of the podcast

  This should be easy to hit.

  Timeframe: 12 months Goal: Publish 1 episode per week Difficulty: Easy Priority: High

• [New] Get proficient in Rust

  I want to “master” Rust. I want to read 10 books on Rust and implement the code from the books. I also want to make one PR to a Rust library (probably something in data e.g. Lance, Datafusion, etc. or AI)

  Books for me are my preferred way to learn and to get into the details. But implementing is the only way to make it stick.

  Timeframe: 12 months Goal: Read 1 book per month, make 1 PR to a Rust library Difficulty: Medium Priority: Medium

• [Renewed] Get into a consistent workout routine

  I am at the moment doing around 3 runs a week and 3-4 shorter workouts. I want to hit 5 runs a week and 5-6 shorter 15 minute workouts.

  Timeframe: 12 months Goal: 5 runs a week and 5-6 shorter 15 minute workouts Difficulty: Medium Priority: High

• [New] Write technical blog posts

  I want to hit 1 technical blog post a month, 1 more practice or process oriented post a week and 1 today-I-learned post a day (excluding weekends). I want to default to sharing my learnings.

  Timeframe: 12 months Goal: 1 technical post a month, 1 practice or process post a week and 1 today-I-learned post a day (excluding weekends) Difficulty: Medium Priority: Low-Medium

• [Renewed] Network with 10 of my idols

  Since I already met 4 of the 10 people, I am reducing the timeline to meeting the rest next year.

  Timeframe: 12 months Goal: Meet 6 more people Difficulty: Medium Priority: Low-Medium

• [New] Launch two more ventures

  I want to launch two more ventures in 2025. I want to hit 3 ventures in total. Each doing over 5k MRR.

  Timeframe: 12 months Goal: Launch 2 ventures Difficulty: High Priority: Mid

• [New] Find three perfect clients for the agency

  The agency is already running and doing well. At the moment, it’s a lot of project work. I want to shift to a more long-term engagement with SMBs or enterprises, who want to kick start their AI journey from scratch (near zero AI experience).

  The goal is not to do a lot of implementation work, but to advise them strategically on how to put the systems, people, processes, and tools in place to make AI work for them. I want to be technology-agnostic. I want to solve problems. Whether that’s through a custom implementation or through finding a SaaS solution that solves their problem.

  Only if it is really the best path, our team will do the implementation, but while training their internal teams to do it themselves in the future.

  I want to find three such clients in 2025 and start the engagement with them.

  Timeframe: 12 months Goal: Find 3 clients Difficulty: High Priority: High

• [New] Launch a new podcast series

  I want to launch a new podcast series focusing more on the implementation side. Either where I am doing the implementation or where I am getting an expert walkthrough a demo or doing live coding.

  Timeframe: 12 months Goal: Publish bi-weekly episodes of the new podcast series Difficulty: Medium Priority: Low-Medium

• [New] Build personal brand

  Reach 5k followers on LinkedIn, grow newsletter to 100 subscribers.

  I have tried a lot around with posting on LinkedIn. I challenged myself to post daily. But too be honest, I am not happy with the quality of the posts. I want to focus on quality over quantity. Post less frequently, but focus where I can actually add value. Where the content cannot be generated by AI. So no slop.

  This will be:

  • Tips and tricks on implementations
  • Sharing code and examples from the ventures and agency
  • Sharing research from the podcast
  • Sharing patterns for building AI systems
  • Sharing learnings from the ventures and agency

  So only hands-on stuff. No fluff. No news. No research papers. No AI generated content.

  I want to become better at communication. So I won’t use AI to write. I will write it myself.

  Timeframe: 12 months Goal: 10k followers on LinkedIn, 100 subscribers on newsletter Difficulty: High Priority: Medium

• [New] Spend more time with my family

  I want to spend more time with my family, my girlfriend, and my really close friends. My social life has been a bit neglected this year.

  I don’t really mind going “monk mode” for a few months. But it’s getting harder and harder to get out of it once you are in it for a while.

  Timeframe: 12 months Goal: Spend 1 weekend a month focusing on social life. Difficulty: High Priority: High

2024 in Review: What I Learned

When I start a something new, my first year is mostly about learning. I don’t know anything yet, I don’t even know how little I know. I can’t optimize shit, because I don’t know where to start.

I ramped up the agency with my cofounders. We started a venture studio.

I started the podcast.

I have started to move my own consulting into more of a strategic role. I advise on how to find AI use-cases, how to build AI systems, and how to put processes and practices in place to make AI work for companies.

It seems like a lot of work in parallel. But there are a lot of feedback loops between the different projects. The podcast helps with advertising the agency. My own consulting helps me stay close to the market and the technology.

So everything is working together. Forcing me to become better at building AI systems.

The thing I underestimated is all the little, hidden tasks you have to do to make something real. For the podcast, the recording, publishing, editing, marketing, but also figuring out the right topics, and the right guests. For the agency, the amount of time spent with contracts, accounting, onboarding and training new employees.

Each project has its own set of tasks. Each task with its own learning curve.

I think I would have been better off starting the venture studio a bit later, so I could have limited the amount of operational overhead. The little stuff adds up.

I know in the future this will pay off, but at the same time, I had to learn a lot already for ongoing projects in the agency and scaling the business overall, so the venture studio felt like a stretch.

In the agency, I did a lot of “keyboard” consulting. I did the implementation (some AI system), handed it over, and then moved on to the next project

“Keyboard” consulting does not have the impact I want. It doesn’t align the incentives of the clients and my agency.

I want to shift the agency to a more long-term engagement with SMBs or enterprises, who want to kick start their AI journey from scratch (near zero AI experience).

It’s not just about the implementation. It’s about putting the systems, people, processes, and tools in place to make AI work for them. And only if they need it, our team will do the implementation, but while training their internal teams to do it themselves in the future.

For this to work, I need to niche down.

I want to focus on SMBs, who want to get started in generative AI, and are in document-heavy industries, where the document processing is part of their core business (e.g. insurance, banking,…), but where compound AI systems are necessary (combining generative AI, search, ML, etc.)

Why this niche? It’s where I have the most experience in. I think the future will look more like “AIOps”. You move from department to department and talk to the people to figure out what takes too much time, where they are losing revenue or need to cut costs. And then you figure out how to use technology to solve their problems. I don’t want to be hired to build, but to solve problems. Whether that’s through a custom implementation or through finding a SaaS solution that solves their problem. I want to be completely independent on how we solve their problems.

At the moment, my co-founders and I have different clients with different needs and from different industries. We already distribute the work across the partners. But I want to niche down in a way, so each partner can specialize in their area of expertise and can focus on a specific industry.

The venture studio started by accident. It emerged. Clients didn’t just want solutions - they wanted to build products together

We do the implementation and they bring the market access and problem space. They basically get to cut the financial risk of the venture and we can participate in the upside.

So far, it’s going great. We are selling one venture at the moment and scaling another one to a few thousand MRR.

It pays to be close to the customer. Every feature we built came from real needs. We learned fast because we had users from day one. No guessing what might work. Features matter less than workflows. UI polish less than actual use. We cut scope to what users need today, not what they might want tomorrow.

It pays to not strive to far from existing solutions. In one of the venture, we re-imagined how M&A advisors work in a data and AI-first world. But, we re-imagined too much in the beginning. We had a new interface, a new workflow, and a different pricing model. You cannot reinvent too much at once. We went back and simplified. Rethought on what we really want to re-imagine and what we want to keep as close to the existing solutions. It makes it easier to adopt and easier to sell. We are like (existing solution) but (better|faster|cheaper|…).

I plan to turn this into something bigger. There are multiple paths we can follow:

1. We continue our coop-model, where existing clients can bring their own ideas and we build them.
2. With the agency, we have to luxury of seeing a lot of problems and solutions. We can start to look for common problems across implementations or industries and build a product around that.
3. I have one more idea, but I am not ready to tease it yet.

I will share more about this in 2025.

I started the podcast to talk to interesting people and learn from them.

Most people don’t know that it’s actually my third podcast.

The first two failed miserably.

This one is taking off.

Too be honest, I did not expect the amount of listeners and engagement.

In most things, it takes multiple tries to get it right.

I have gotten a few partnership and sponsorships offers, which I have declined. I am not sure whether it will be something I want to pursue in the future.

I like to stay independent. I can have on who I want, I do not have to answer to anyone, and the guests can share their opinions without any pressure or restrictions.

The only thing that I have seen so far that would be interesting is to do paid interviews (like Machine Learning Street Talk). Basically, the sponsor pays for a few of their engineers to be interviewed, but they do not control the content.

I will scale up the podcast in 2025. I will have a more consistent schedule, more guests, and more episodes.

So far, I have only done interviews.

I want to share more of my learnings from running the agency and the venture studio (and the ventures itself). This will be the focus of a new podcast series, which is much more hands-on.

The two series will complement each other, but be independent.

• In the interview series, the guest can share their experience.
• In the hands-on series, I will share actual code and examples from the preparation for the interviews, from the guests, or from my ventures / agency. Or bring on other guests to do live coding.

The main thing I learned is that I have to make it very easy on myself to be consistent. Plan ahead. Ask people for help. Use tools to make it easier. Automate repetitive, consistent tasks.

When I can sit down and do the same thing for multiple hours, I can get a lot done. Do any work (e.g. recording, editing, publishing) in batches or sprints. Then rest. I also learn faster. I make a mistake, learn from it, and then move on.

I will kickoff the next season (MLOps) in Feburary. The beauty of MLOps is that the hands-on part can be anything. So I can use the code from the ventures or the agency and also get other people on that I do not interview in the MLOps series.

I am not happy with my personal finances yet or rather the process I have in place to manage them.

I am managing multiple accounts, each with it’s own bank account, credit card, and bookkeeping.

I want to keep it this way to simplify my taxes, but I need to find a way to consolidate the information and make it easier to manage.

I also have not found a great way to manage invoices and bills.

In the agency and in my own bank accounts, I should track the invoices with the transactions, but it takes a lot of time to get all the invoices out of the different tools. Especially, usage-based bills.

It pays off to set larger limits e.g. on OpenAI, especially in the EU, because you pay roughly 0.36 cents for each transaction in Dollar. This might not seem much. But we had at least 100 in the last few months, which cost us 36€. And that’s just for one service.

I want to try out Ramp, but didn’t find the time yet.

I really regressed in the last year. It’s the first time in the last 10 years that my fitness level is worse than the previous year.

I need to find a better routine and a way to fit it into my schedule.

I think I have figured out a way to make it easy for me to be more consistent.

I want to do 4 runs a week and 5-6 short 15 minute workouts.

I have a good base when it comes to muscle and strength. I want to lose mass, get injury-proof, and improve endurance.

I want to train for dynamic ranges of motion in various positions, like an ‘all-terrain vehicle’. So full-body movements with a lot of range of motion with lots of different angles and variations. Think: Lunges in all planes. Turkish get-ups. Single-leg moves. This is not the best to build muscle, strength, or endurance. This also isn’t about getting ready for a specific sport. It’s about stressing the body in weird positions.

The runs will be mostly base runs. 10% at 90% intensity, 5% at 95% intensity, and only 5% at maximum 100% effort. So one Norwegian 4x4 a week and the rest easy runs, where 1 or two might end with a few sprints.

So: Consistent training over always pushing for high intensity. Adapt training based on current state; when I feel good, push harder. When not, do less or some Yoga.

Looking Forward

Three main themes emerge for 2025:

1. Build for the long-term: Whether it’s the agency, ventures, or content creation, I’m shifting to building slow and setting everything up to be consistent and deliver.
2. Health first: Do the basics. I am not looking to be the best runner, to be the strongest, but to be injury-proof and healthy for the long-term.
3. Share more, build in public: I want to document and share the journey, both successes and failures.

Let’s do it!