Nova Blog - A year without writing a single line of code: how Nova's Head of AI actually works

From data science to agents: riding every AI wave at 31

Jesús Copado grew up in Lanzarote, enrolled in aerospace engineering in Madrid — "I went for the hardest degree because maths came naturally to me" — and realised halfway through that planes didn't interest him as much as software did. He pivoted, did an Erasmus in Germany, completed a master's in autonomous systems split between Finland and Budapest, and spent several years at a deep learning startup in Munich right before ChatGPT arrived.

When the world saw GPT-3.5 in November 2022, Jesús had already been working with computer vision neural networks for months. "It was a week without sleep. Now it's been two months without sleep." His trajectory is an almost perfect timeline of the names the field has cycled through: data science → machine learning → deep learning → agents.

Today, at 31, he is Head of AI at Nova, leading the Data & AI team responsible for the Nova Recruiter search engine — now processing 800 million profiles — and for the agentic layer that is transforming how the entire company works.

"I started with data science, which is what it was called back then. Then machine learning, deep learning, and now agents. I've been in AI in every form it's taken."

— Jesús Copado, Head of AI · Nova

A year without opening an editor: what working "agent-first" actually means

The line that opens the episode sounds almost provocative to any engineer: Jesús hasn't written a for loop in over a year. The nuance matters: the machine codes, he orchestrates. Since adopting Claude Code as his primary tool roughly a year ago, his relationship with code changed fundamentally.

The workflow is always the same: natural language (or voice) → instruction to the agent → the agent plans, researches, writes, tests, and opens the merge request. "What used to take 20 hours now takes 20 or 30 minutes to configure." In the past two months, with the arrival of Claude Opus 4.5/4.6 and Codex 5.2, the leap has gone further still: there are now atomic tasks for which he doesn't even review the output.

The difference from coding assistants six months ago isn't just quality. It's architectural: this is no longer a model suggesting the next line in an IDE — it's an orchestrator agent that launches web searches, runs tests, interacts with GitLab, and updates the ticket in Linear, all within the same conversation.

The 100x engineer stack in 2026

For anyone who wants to replicate this way of working, Jesús identifies four essential layers:

Jesús Copado's recommended stack (2026)

Layer 1

A terminal agent (CLI) — Claude Code, Codex CLI or Cursor Agent. The choice matters less than the habit: the key is making the conversation — not the file — the unit of work.

Layer 2

Voice dictation — Jesús uses Whisper mapped to his Function key. You speak 2–3× faster than you type, and messy formatting doesn't matter: the model distils the intent. "It lets me ramble. I think out loud."

Layer 3

Connected MCPs — Model Context Protocols wired to the tools you already use: Linear (tickets), Notion (docs), GitLab/GitHub (repos), Gmail/Slack. The agent acts directly inside each one without you switching windows.

Layer 4

A skills folder in Markdown — The real differentiator. Plain text files that give the agent the memory and working style it would otherwise need you to repeat every session.

"The core piece is a terminal agent. From there you have internet access, you can connect MCPs, and the user experience ends up feeling very similar to ChatGPT — you talk to it in plain language and it replies. But what it can actually do is radically different."

— Jesús Copado

Open Claude: the agent that lives in Slack and never sleeps

Everything described so far still requires the human to start the conversation. Open Claude — the open-source framework built by Austrian developer Peter Staber, recently hired by OpenAI — takes things a step further: the agent listens proactively.

What is Open Claude

A system that deploys an AI agent inside any interface (Slack, WhatsApp, Discord, Telegram) with three key context files:

user.md — who you are, your personal and professional context
soul.md — the agent's "soul": its personality, values, decision-making style
instructions.md — what it should do and how

The difference from Claude Code in the terminal: Open Claude is permanently active, receives inputs without being called, and can act on them autonomously or with minimal human sign-off.

Nova already has a bot like this living in a Slack channel. When a headhunter reports a bug with a screenshot — "this isn't working" — the agent sees it, understands the problem, creates the task in Linear, and, if it's a simple bugfix, resolves it without any engineer ever getting involved. "What Open Claude gives you is an agent that's always alive, always watching."

The near-term vision: someone from the business team reports a bug in Slack, and ten minutes later it's fixed in production without a single human touching a line of code.

Beyond software: what Nova is already doing with AI today

Jesús is emphatic that none of this is just for engineers. The concrete examples at Nova go well beyond code:

Democratised data analysis. Previously, an ad hoc database query required knowing SQL or waiting for someone on the data team to run it. Now anyone in the company talks to an agent in plain English that translates on the fly to SQL, runs the query, and returns the result. "How many Nova members signed up for Gravity this week?" gets an answer in seconds.

Candidate reports for headhunters. Nova's talent team presents candidates to clients with detailed write-ups. It used to be manual: gather interview notes, fill out a form, write the summary. Now the headhunter tells Claude "present this candidate for this role" via MCP with Nova Recruiter, and the report lands directly in the client's inbox. "That whole tedious form is gone."

Content and comms. LinkedIn posts from video files, recurring board KPI emails, inbox triage and categorisation. Jesús mentions that this morning the agent cleared 300 stuck emails: archived them, categorised them, and drafted replies for him to review before sending.

The future of the labour market: cautious optimism, universal basic income, and an honest "I don't know"

The second half of the episode moves into uncomfortable territory. Ramón puts the question directly: if all knowledge work is automatable, where does that leave us?

Jesús describes himself as "a little pessimistic" and makes an argument that rarely gets articulated this clearly: in every previous technological revolution, when one type of work disappeared another emerged. Dock workers were displaced by mechanisation, but accountants appeared. Accountants could have been wiped out by Excel, but there are more than ever. The pattern was always the same: from physical to intellectual. The problem with AI is that it automates the intellectual, and there is no obvious next rung on the ladder.

What worries him most isn't the change itself but its speed. "There are people right now in knowledge-work jobs whose work is already pointless. I wonder whether they'll be able to relearn everything." His view on universal basic income is clear, even if he knows it's unpopular in some circles: "if we don't get there, it's going to be very difficult not to end up with severe social tension."

"I don't know if it's the change itself or the speed of it. That's what worries me. We're heading towards a world where we'll have to revisit our social systems."

— Jesús Copado

The second brain: Obsidian as an AI-powered idea network

Outside what he builds at Nova, Jesús is investing significant personal time in something he describes as a "second brain": an Obsidian vault where his notes — books, films, articles, projects, ideas — connect to each other like Wikipedia.

The difference from Notion or Apple Notes isn't just technical (files live locally as Markdowns, not in the cloud). It's philosophical: there is no imposed folder hierarchy. An idea goes in, gets linked to other notes, and the graph grows organically. AI enters at two points: to distil voice captures into clean notes, and to surface connections between existing notes that the user wouldn't have spotted themselves.

"I'm using Obsidian alongside Claude to build a small blog where I want to bring in more philosophy, more social reflection. I think it's the right moment to think out loud about what's happening." The system lets him capture an idea via voice while walking, forget about filing it, and find it later already connected to a book he read three months ago and a conversation from the previous podcast.

From Lanzarote to Nova via Accenture, Munich, and a personal brand

Jesús's career arc is that of someone who learns primarily through contrast: Accenture taught him what he didn't want (large organisations where individual impact is imperceptible). The Munich startup confirmed he wanted agility. The personal brand — the YouTube channel he started building as a side project — is what eventually put him on Nova's radar.

The call from Cristian (Nova's CPO) was direct: "Hey, I'm watching your content. Do you know anyone for a Head of AI role?" What convinced Jesús wasn't the salary or the title but the promise of building something from scratch — the kind of problem that activates his abstract thinking — inside a team with real ambition.

"The decision to join was very human. There was an energy in the conversation that said: we understand each other, we don't know exactly where we're going, but we're going together." That is, ironically, precisely the kind of value he argues AI cannot yet replicate.

🔑 Key takeaways

1. Agent-first is a mindset, not a tool. The real shift isn't adopting Claude Code or Codex — it's redesigning your workflow so the conversation, not the file, is the unit of work.

2. Context is the differentiator. Two people with the same model get radically different results depending on the quality of context they inject. Markdown skills files are the most practical way to persist that context.

3. Voice dictation multiplies thinking speed. You speak 2–3× faster than you type, and informal formatting doesn't matter — the model distils intent. It's especially useful for giving rich context without friction.

4. Open Claude is the next step: proactive agents. The difference between an agent that waits for your command and one that listens to Slack continuously is enormous for product teams.

5. Junior roles are the most exposed, for now. The labour market is already feeling it in software. The speed of change is the real risk — not the change itself.

6. Personal brand is still the best ROI on time. It's what turned Jesús from "a good engineer" into "the person Nova specifically wanted." In a world where hard skills commoditise, visibility compounds.

‍

Nova: where the best tech talent connects

Nova connects 25,000+ high-potential people across 80+ countries. If you want to surround yourself with people at the frontier of what's happening in AI and technology, this is the place.

→Apply to join

‍

A year without writing a single line of code: how Nova's Head of AI actually works