# Acelerando #01: André Thiessen on how Picnic works with AI

> In the Acelerando pilot, João Ferreira and André Thiessen talk about coding with AI, support 75% resolved by AI, an automated blog and the Claudinho agent.

Source: https://usepicnic.com/en/blog/acelerando-01-andre-thiessen-ai-at-picnic · Published 2026-11-19

**Acelerando is Picnic's podcast of conversations with people who use AI for real at work.** In episode 01, the pilot, João Ferreira, Picnic's co-founder and CEO, talks with André Thiessen, who has been at Picnic for almost four years and looks after product and engineering. The episode is in Portuguese, with Portuguese subtitles on YouTube.

▶️ [Watch on YouTube](https://youtu.be/F2m7J-IyLuk) · 🎧 [Listen on Spotify](https://open.spotify.com/episode/03UhtRGoBMtkqKswuUb4yI)

[Vídeo: Engenheiro do ITA: “não existe desenvolvedor front-end” | Acelerando #01](https://www.youtube.com/watch?v=F2m7J-IyLuk)

It is 38 minutes on how Thiessen uses AI to code, why he thinks "there is no such thing as a front-end developer" and what we did to be less afraid of letting AI work on its own. We also open up how Picnic runs: customer support, this blog and Claudinho, the agent that lives in our Slack. Below is a summary that follows the episode's chapters, with the timestamp of each so you can jump straight to it. Quotes are kept in the original Portuguese, with our translation underneath.

## Opening

[0:00 on YouTube](https://youtu.be/F2m7J-IyLuk)

João explains the idea behind the show: bring in people who use AI every day, far from shallow content that celebrates uninteresting results and steamrolls processes full of nuance.

> "E a ideia, a premissa é como é que você constrói uma empresa na era da IA."
>
> (And the idea, the premise, is how you build a company in the age of AI.)

## Thiessen's path to Picnic

[2:03 on YouTube](https://youtu.be/F2m7J-IyLuk?t=123)

Thiessen left Brasília to study at ITA, a Brazilian engineering school. In his second year he started a software engineering internship, almost all front-end, and spent days changing something about a button on a screen. Around the same time he went to Ethereum Rio knowing nothing about blockchain, met João and, with friends, was among the winners of the event's hackathon. At the end of the year he was one of the DevCon scholars in Bogotá, ran into the Picnic team again, quit the internship and, some time later, joined Picnic.

He says he arrived "very green" at programming, when ChatGPT was just getting started, and that all of his professional growth happened here. What drives him was never a programming career:

> "o que mais brilhava nos meus olhos era construir, enfim, entregar entregar algo, construir fazer parte"
>
> (what really made my eyes shine was building, delivering something, being part of it)

## Learning with AI: from tutor to delegating

[7:05 on YouTube](https://youtu.be/F2m7J-IyLuk?t=425)

Thiessen's use of AI went through three phases. Before AI, he had to hunt for sources and study until he knew enough to build, and it was slow. Early on, AI was a chatbot: he pasted a snippet to understand what that code did. Then came the phase of learning faster what he needed to build. Today he often does not need to learn the details at all. For the prototype of his ITA graduation project, for instance, he worked at the high level and did not write the contract by hand.

> "Tem um processo ali meio que de confiança, quanto mais você confia, você consegue delegar para a AI e ir executando."
>
> (There is a kind of trust process: the more you trust, the more you can delegate to the AI and keep executing.)

> "E aí, uma vez que você tem o alto nível, consegue ter o direcionamento, entender o que está acontecendo, muitas vezes sem os detalhes, você voa, basicamente."
>
> (Once you have the high level, you can give direction and understand what is going on, often without the details, and you basically fly.)

## Why some people use AI better than others

[10:05 on YouTube](https://youtu.be/F2m7J-IyLuk?t=605)

For Thiessen, the difference is critical thinking: understanding what the AI needs from you, what your role in the flow is, and judging what it delivers. Mastering a language's syntax is no longer the point. What matters now is looking at an interface change and checking that it respects the user experience, the design system and the color palette.

> "Eu acho que não existe desenvolvedor front-end basicamente e a pessoa tem que ser um desenvolvedor de produto."
>
> (I think there is basically no such thing as a front-end developer; you have to be a product developer.)

And where many people go wrong:

> "onde muitas pessoas muitas vezes erram é em pular essa etapa de pensamento crítico"
>
> (where many people often go wrong is skipping that critical-thinking step)

## Those who resist and those who just mark it done

[15:16 on YouTube](https://youtu.be/F2m7J-IyLuk?t=916)

João adds adaptability and tells a story about one extreme: a professional whose company required everyone to use AI, and who started asking Claude to do all of his work, without reviewing it, marking everything as done. He used the tool, but skipped the part about making it work better than a bare request.

Thiessen describes the other extreme: people with long, very senior careers who struggle to change, sometimes for an almost moral reason.

> "nossa, estou abrindo mão do que eu fiz durante 20 anos para que uma AI faça"
>
> (wow, I'm giving up what I did for 20 years so that an AI can do it)

He also recalls a question he saw on Twitter asking people for their favorite programming language:

> "qual a sua linguagem favorita de programação hoje? Falei, pô, Claude, como assim?"
>
> (what's your favorite programming language today? I said, come on, Claude, what do you mean?)

## Uncle Bob, security and competing without AI

[17:46 on YouTube](https://youtu.be/F2m7J-IyLuk?t=1066)

João tells the story of Uncle Bob, who lived through the move from Assembly to C as a professional programmer. At first his team hand-optimized the Assembly the compiler produced. When AI arrived, as João tells it, he decided not to repeat that mistake: he stopped reading code and works through his tests and his structure.

For João, there are areas where not using AI is already a risk:

> "se você não estiver usando IA para se defender você está muito enrolado porque o cara está usando IA para te atacar"
>
> (if you are not using AI to defend yourself, you are in deep trouble, because the other guy is using AI to attack you)

Thiessen agrees. In company processes, whoever works without AI loses speed and effectiveness. In security, they may lose the company.

> "Então, você querer competir de forma humana ainda com AI, é muito complicado."
>
> (So trying to compete as a human against AI is very hard.)

## AI customer support: 75% and 95%

[19:39 on YouTube](https://youtu.be/F2m7J-IyLuk?t=1179)

João gives Picnic's own example: in some functions, the best quality available today already comes with AI. In customer support:

> "75% de tudo que a gente recebe de atendimento é resolvido de ponta a ponta com a IA"
>
> (75% of everything we receive in support is resolved end to end by AI)

The AI escalates cases to partners, understands blockchain transactions and digs deep into each problem. The flow also improves itself: when a customer rates the service badly, or when the AI cannot solve a ticket, it reads that ticket, works out what it could have done better and adjusts.

> "A gente teve, nessas últimas duas semanas, 95% de atendimentos positivos."
>
> (Over these last two weeks we had 95% positive support ratings.)

Those are the two weeks before the recording. For comparison, when the team was human the figure was around 60% to 65%.

> "vai ter um ponto que você vai ter a melhor qualidade naquilo que você faz com a IA."
>
> (there will come a point where the best quality in what you do comes with AI.)

## Automated blog and Claudinho

[22:10 on YouTube](https://youtu.be/F2m7J-IyLuk?t=1330)

Thiessen picks two projects he is proud of. The first is the articles and SEO work on this blog, which he had tried to run before without being able to give it enough focus. With better models and a lot of iteration, we started publishing articles that help people answer their questions and discover Picnic, and visits to the site grew.

> "esse processo hoje é praticamente 99% automatizado por AI. Eu acho que é só o clique final que eu dou nos merges ali."
>
> (today that process is practically 99% automated by AI. I think the only thing left is the final click I give on the merges.)

The second is Claudinho, which Pury created:

> "o Claudinho nada mais é do que um agente da Picnic, com todo o contexto do Picnic"
>
> (Claudinho is simply a Picnic agent, with all of Picnic's context)

It lives in Picnic's Slack, reads any thread, sends announcements and job alerts. Anyone can open a thread on one of its alerts, which already carries the context. That lets different areas work together without a call: Isabela, from legal, opens a PR with Claudinho from one of its alerts, and a reliability alert turns into an investigation on the spot, while the context is still fresh.

> "E, opa, investiga isso aí para mim?"
>
> (And, hey, can you look into this for me?)

## Big company vs. small company

[27:30 on YouTube](https://youtu.be/F2m7J-IyLuk?t=1650)

João notes that this kind of Slack integration has become more common but has not reached everyone yet, and that big companies resist more. For Thiessen, the problem is context:

> "Eu acho que a empresa grande vai sofrer muito com qual contexto eu vou liberar, para quem eu vou liberar cada contexto."
>
> (I think big companies will struggle a lot with which context to open up and to whom.)

In a smaller company this is easier to organize. João adds that big companies carry a large cost structure and will have a harder time adapting, but also a very strong incentive to do it.

## How to stop being afraid of letting AI loose

[28:16 on YouTube](https://youtu.be/F2m7J-IyLuk?t=1696)

João asks Thiessen for his turning points. The first is being willing to change how you work: he was slow to move to Cursor and later resisted trying Claude Code. The second:

> "entender o que você precisa pra ter menos medo de deixar a AI tomar mais conta"
>
> (understanding what you need in order to be less afraid of letting the AI take over more)

At first he reviewed, line by line, the code the AI wrote. Today that would be impossible:

> "se eu revisasse o código linha por linha até hoje eu acho que eu ia gastar todos os meus dias só revisando os códigos do João"
>
> (if I still reviewed code line by line, I think I'd spend all my days just reviewing João's code)

The way out was to have the AI review the code, iterate many times and write review skills with what he himself would look for. He warns that simply not being afraid leads to the mistake of jumping straight to the result. What works is knowing what you need in order to trust.

## AI testing its own work

[31:47 on YouTube](https://youtu.be/F2m7J-IyLuk?t=1907)

Trust comes from the AI being able to test everything. First came unit tests, which the app now has plenty of. Now we are building our own sandbox, where the AI runs the product and tests, for example, a change to the Pix flow against the provider's sandbox, on both back end and front end.

Thiessen says he created a design in an artifact with Claude and asked Fable to implement it in the app. It tested in the sandbox, noticed the first version did not match the design and fixed it by itself, without him having to do the QA.

> "Dado que ela consegue testar, eu consigo começar a ter mais confiança de soltar."
>
> (Given that it can test, I can start trusting it enough to let it go.)

> "uma vez que você ganha essa confiança velocidade consegue escalar infinito."
>
> (once you gain that trust, speed can scale infinitely.)

## The bug that only broke on Safari

[35:21 on YouTube](https://youtu.be/F2m7J-IyLuk?t=2121)

João shares a recent case: a feature with video upload and a 4 MB limit per video. Testing in the sandbox, the AI recorded a video, could not get it under 4 MB and found another way to upload the file on its own. And the case was very specific: it worked on Chrome but not on Safari, one of those problems that only shows up sometimes.

> "Assim, eu pedi pra ele testar, mas foi, tipo, literalmente, testa aí."
>
> (I mean, I asked it to test, but it was literally just "test it".)

Thiessen points out that before, we almost never tested a feature across many phone and browser models. With the AI testing, you can ask for ten different models.

> "Talvez demore o tempo e tal, mas vai ser delegado, né?"
>
> (It may take a while, but it will be delegated, right?)

## Closing

[37:33 on YouTube](https://youtu.be/F2m7J-IyLuk?t=2253)

> "a gente tá em tempos bem legais aí e com certeza tem muita coisa pra fazer ainda"
>
> (these are really cool times, and there is certainly a lot still to do)

João thanks Thiessen for his partnership in building Picnic, an easy-to-use [self-custody wallet](https://usepicnic.com/en/blog/what-is-self-custody) that works as a global account, and asks listeners for feedback.

## Watch the full episode

Acelerando is a Picnic podcast of conversations with people who use AI for real at work. This is the pilot, so feedback helps a lot: tell us in the YouTube comments what you liked and what we should change next time.

- ▶️ [Watch on YouTube](https://youtu.be/F2m7J-IyLuk)
- 🎧 [Listen on Spotify](https://open.spotify.com/episode/03UhtRGoBMtkqKswuUb4yI) and [follow Acelerando](https://open.spotify.com/show/1o8kB80DWR0t728ZGPi74q)

If you want to know the product we build, Picnic is at [usepicnic.com](https://usepicnic.com).
