Staying Human with AI

First of all, I have to say I'm excited to start publishing on Fika, a tool by a friend of mine, Pau, who believes a different Internet is possible, and he's proving it with tools like Fika, which feel so unique in a tech world that gravitates towards pattern homogeneity. I'm currently using Ghost, which I'm a happy subscriber of, but I'd rather support friends, especially when they are trying to build something different.

I thought I'd write a first blog post talking about how I use AI these days, something that keeps changing, and that has nothing to do with how I used it months ago. For context, I'm a builder turned founder and I'm now the CEO of Tuist, a productivity toolchain for app developers, and we are a small company of 4 people. Because of that, my responsibilities span across several domains: doing sales calls, writing content for marketing, doing some support, fixing small issues and so on… Because I find a strong joy in building, I also allocate some time every day to do some building that brings me joy and balances the other tasks/responsibilities that are less fun to navigate.

Coding

My coding these days divides into two categories. On the one hand, I tackle small issues or bugs that I notice arise. I try not to take on big projects because the frequent context switching the company state requires would put myself in someone else's critical path, or drag a project forever. Also, my ability to go deep into a particular problem is very limited due to the context switching, and this is something where agentic coding tools come quite handy. First, I use tools like Codex & Claude Code to go deep into understanding a particular issue or bug, and have a conversation about what the best solution might look like. I treat it as a pairing partner, and we focus initially on understanding the trade-offs of all possible solutions. This is a lot of fun, because they do the exercise of going deep into the codebase and the problem, and then I can leverage my expertise (or gain extra) to guide it to the best path. That work usually results in a small PR, which I usually double-check, but at times I miss details because of a bit of ADHD (I'm already thinking about the next priority).

The second piece of coding work that I do is more creative and long-term. I enjoy thinking how Tuist could evolve in 1, 2, and 5 years from now, so I prototype ideas. This is the funniest part, and it reminds me a lot of the beginning of Tuist. It's a white canvas. When working on the day-to-day responsibilities and priorities, the work is very constrained. The immediate priorities are reaching profitability, within the Apple ecosystem where we have the moat today, and while respecting the mental models that people have built around what you are, even though in your mind you have ideas for what it could be to unlock new business opportunities. This is frustrating, and I hope to talk about this in future blog posts.

And wow, AI is so amazing to do these explorations. I like to read what's happening around us, and in other ecosystems, and I capture the ideas that I believe are good, and the ones that I believe are bad ones, and then have very interesting conversations with LLMs and agentic coding tools about those and then we shape a solution together. When I'm doing this, the programming language is an implementation detail. I know which language would be the most suitable for the solution, so I suggest that early, but once it starts coding, I just focus on the architectural and DX & testing patterns that I believe are good to embrace, and don't care too much about the style of the code. I check the tests though to ensure they capture well what I expect the software to do.

Oh, and I forgot to mention that I also use coding agents to learn about new codebases or contribute fixes to them without having to spend days gathering enough context to make a contribution.

It’d be more accurate to say that I care about the styling. It’s just that I believe that this is something that toolchains have finally solved, so I’d delegate that task to tools like cargo fmt or mix format

Sales

One of my responsibilities consists of turning prospects into customers. This involves having frequent conversations with customers, initially over email, and eventually over Slack, which we usually connect through. In those first emails, I want to make sure the message that I'm trying to send sounds well on the receiving end, and also takes into account cultural aspects that are part of the sales process. For that I draft an email myself, and then lean on LLMs to help me iterate on it grammatically, ensuring the message is well captured, and that will evoke an action on the other end. I believe a well written message can make a huge difference. I like to write those myself because when they are written by an LLM, they don't have my personality, and this is something that I don't like.

If you wonder, I copy the email to the desktop app to the LLM, ChatGPT or Claude, and then copy the iterated message back to the field. I'm not using those agentic experiences built into browsers. Perhaps I will at some point? I got so used to the back and forth, that the pattern solidified and now I don't see much value in changing it. I quite like the jump.

I wish more CRM tools used AI to assist in the sales process. We are excited about operate.so but it's still work in progress. We are currently using Attio, but I think they are missing the opportunity to rethink the solution with AI at the core. Sure, they sprinkled it here and there, but the needs that I have, which I think AI would solve very nicely, remain unaddressed. Some examples:

• Reminders about when it's a good time to do a follow-up, and what some action items remain unaddressed.

• Some important notes that one should know when jumping on the next call with a lead.

• Automatic reminders when it's a good time to start talking about the renewals.

• Identifying the different roles in the conversations that are happening across all the channels: emails, calls, Slack.

Based on what we've seen, I think CRM tools will get there, and I couldn't be happier. I feel sales is an area where a lot could be automated with AI. I believe a human needs to be in the loop, and needs to be the founders there because we bring the technical credibility, but everything else can be automated. I'm seeing a lot of these sales tools that automate outbound emails, or demos, or sales calls, but I don't know… it gives me "Black Mirror" vibes and I have my doubts this will work with developers. We are humans, and commerce is human too, especially in B2B deals like these ones we are doing.

Content

A lot of the marketing of Tuist is done through content where we share not only product updates but also talk about how we see our space, and the vision that we have for it. I wish I did more of this, but as I mentioned earlier, content writing suffers, like deep coding sessions, because my work these days entails a lot of context switching. Most of my blog posts start with iA Writer on my iPhone, some are even completed there and then I do an editing on desktop and publish. Other times I just use iA Writer as a place where I jot down ideas, and then later on, when I have a quiet moment, I sit down, and then turn those ideas into a blog post.

Where do I use AI, you might ask? I do so when reviewing the grammar of my blog post. I used to use Grammarly for that, but I've replaced it with LLMs now. I'm very clear about not changing the style or the personality, and just focus on the typos that I might have introduced, since I'm not a native English speaker, so there are many tiny things that I miss without noticing it. I tried to write some blog posts with AI, doing a bunch of iterations, but I ended up giving up in the end because it doesn't sound "me". I really enjoy when there's a personality in the content that someone publishes, like in Pau's posts, and I don't want that to be gone with a productive mindset that AI tools bring to the table.

Research

When there's something I want to learn about, I go and ask LLMs. This has replaced Google completely for me which I use when I know exactly what I'm looking for, which is often not the case in my day-to-day job, where I've heard of a technology, or a concept, and I want it to expand on it and share some references with me so that I keep digging if I need to. I tried to use tools like Obsidian or Logseq to keep track of ideas that I'd learn, and connect them, but I don't feel the need for those tools anymore. Sure, I'm dependent on some LLM having that interconnection of ideas, which is not something I control, and I'm also very reliant on my brain, but so far it's been working fine for me. Jumping between iA Writer, LLMs, and sometimes paper is a setup I feel very comfortable with. In fact, I have to admit I enjoy a bit of chaos when doing this type of work because I believe chaos and creativity go hand in hand. When I try to eliminate the chaos with structure, for example trying to find the right Logseq structure to connect notes, things start to feel less creative and fun.

Reading

I enjoy reading the blog posts that are worth reading, and those are usually something that you can filter out by reading a few sentences. However, there are others that you can tell right away, that those are the "it could have been a meeting" kind of post, so when I come across those, I just go straight into the LLM and ask if there's anything meaty in it worth learning from. Often times the core of the blog post is as shallow as the first lines, but in other cases there might be something substantial hidden in it, so I just get the summary and move on. I try not to do this a lot, because my brain, which tries to optimize where possible, will gravitate to do the same with every piece of context, as if the function of every piece of context was for you to derive something meaningful out of it and not just simply the joy of reading.

I'd say my usage of AI is very basic, compared to what I'm seeing out there, but I'm happy with it. While it's a productivity tool, which helps me focus on the things that bring me the most joy, and automate some of the ones that don't, I don't try to treat it as a tool that should automate everything, including writing, or writing some code here and there, because then everything would revolve around productivity, making everything that I do feel miserable in some way. I don't want everything in my life to be optimized towards productivity. I want things to be done for the joy of doing them, and that's the reason why I don't get obsessed with trying to put AI everywhere.