Barcelona - Itnig - 22nd April. CTO Networking + Tapas + Short talks from CTOs. Apply here.

The way users interact with software is changing. For readers of this newsletter, probably earlier than for most. But in some use cases, this shift is already starting to reach the early majority.

I see product agentification evolving in three stages.

1. Traditional manual products. Most companies are still here. Standard UIs, manual workflows, forms, dashboards, documents, and information spread across different tools.

2. Agent for the product. Many companies are now adding agents as a new interface layer. Sometimes for narrow use cases, like searching docs or drafting text. Sometimes as a broader way to interact with the whole product from places like Slack, Teams, or Discord. Vercel's recent Chat SDK launch points in this direction.

3. Product for the agent. This is where things get really interesting. A smaller group of companies build products that agents can use directly through MCP servers or CLIs. We are already seeing this (specially for devs) with Claude Code plugins, ClawHub skills, and Codex plugins.

The key is shared context. LLMs are only as useful as the context they can access and use effectively. Imagine agents that can:

access the full context of a company, including transactions, conversations, CRM data, and metrics

identify and select the most relevant context for each task. I believe this is the area that needs to improve the most.

use any productive tool, whether that means code, contracts, presentations, email

This does not mean traditional UIs become useless in every case. I still want to open my banking app and see my balance without having to ask an agent. But I do think most companies will be pushed to adapt to this shift in one way or another.

For that to happen, agent inference costs need to make sense, which I believe they already do for many productive use cases. At the same time, the pricing model for many SaaS products will likely have to evolve from seat-based pricing to more usage-based or outcome-based models.

Anthropic announced Glasswing, a coalition between major tech players to use it's unreleased model Claude Mythos to find and fix critical software vulnerabilities before AI-powered attackers can exploit them. The claims the announcement make are pretty crazy:

"Over the past few weeks, we have used Claude Mythos Preview to identify thousands of zero-day vulnerabilities (that is, flaws that were previously unknown to the software's developers), many of them critical, in every major operating system and every major web browser"

Does GitHub still merit "top git platform for AI-native development" status? The article argues that GitHub is buckling under AI-agent-driven load, exposing outages, weak strategic focus, and a failure to evolve into the AI-native developer infrastructure the market now needs.

Interesting piece criticizing OpenAI's acquisition of TBPN. To be honest, this felt a bit sad to me, because TBPN's independence was its greatest strength.

How I Run a Fully-Remote Software Engineering Standup: A concise guide to running short, human, daily standups that surface blockers early and keep everyone aligned.

Discussion on context engineering. @aldo asked how people where managing context for their AI agents. I shared this guide that shows smart strategies for improving context, like externally saving context, rag for pulling relevant info and using subagents with specific context.

@aleix wrote Rethinking Pair Programming in the Age of AI. If seniors replace pairing with AI-driven solo productivity, teams may ship faster in the short term but fail juniors by weakening mentorship.

@sergio wrote Handling unreasonable expectations practical critique of AI hype, arguing that leaders should ground expectations in context, data, and real business outcomes instead of anecdotes and magical thinking.

Inditex Tech is Hiring a Senior Principal Engineer.