WeAreDevelopers Europe 2026 - my takeaways in one sentence

By Xavier Sala, 10 July 2026

My takeaway from WeAreDevelopers Europe 2026 in one sentence:

The software development lifecycle we have been using for the last 20 years has officially changed.

Policies, rules, skills and specs all part of the repository. AI generating code in cadence that we cannot even approve. PRs are being merged even without seeing the diff. We has humans cannot review such amount of code. Spec driven development. The most used programing language is now English. The rise of agentic development, as early as in the ideation phase.

Without a doubt, we are entering into an amazaing new era for the software systems.

However all this makes me think about

  • the associated costs
  • the safety & trust
  • the different speeds where this agentic assembly line is being used at big corporations, start ups, regulated industries, snall teams, etc.
  • how new features are updated even they are part of the repository as specs

Here are some thoughts from few talks at WeAreDevelopers that really got me thinking.

Talk: Building Modern Distributed Systems using Less AI Tokens

by Przemysław Ładyński

  • AI doesn’t replace engineering! Lets be a Software Engineer(ing Coordinator)
  • AI follows either your engineering standards or the Internet average.
  • Knowledge belongs in the repository and is expressed via policies, rules, skills (ie. to create diagrams with draw.io in the repository, etc.) and specs (architecture, code-style, security, user geatures, etc.).
  • How AI understand knowledge you provided can be tested with evals/datasets and graders.
  • Your competitive advantage is no longer creating high quality systems but orchestrating the statistical machines (LLMs) to deliver it in the most efficient manner.
  • Same prompt. Different rules. Different result. -> reduced code volume, inreases token efficiency, PR readability, change locality and context efficiency.
  • Before the AI Revolution.

Talk: Designing for Agents Will Make You Better at Designing for Humans

by Dana Lawson

  • The new most used programming language: English
  • The builder is no longer just a developer.
  • The rise of agentic development - intent as the input and working softeare as the output.
  • Autonomouns development loops - CI/CD evolves from a passive pipeline into a continious, agent-driven feedback loop.
  • Traditional systems exposes endpoints. Agent-native systems exposes intent-level operations that agents can reason about.
  • Request-Response to Event-driven - agents perform best when they can observe system behavior and act autonomously - not poll and wait.
  • Trust & Safety in agentic systems. When agents can deploy code, provision infrastructure, and modify production - trus isn’t optional. It’s the foundation.

Talk: The agentic assembly line

by Thomas Dohmke

  • Jurassic era of the AI coding.
  • More code is generated by AI agent -> increased numnber and changes in each PR.
  • How to review inceased PRs and merge with confidence? by not logging only the diffs but also scores for risks, confidenly, etc.
  • More AI changes are accepted without human review.
  • New SDLC from intent to the outcome by merging without even openning a diff.

Talk: From vibe coding to viable code with spec-driven development

by Julian Wood

  • AI is changing software: from completing development tasks to now proven correct, team-scale and iterative development.
  • Spec-driven development:
    • Persistent - lives in your repo as markdown
    • Iterative
    • Verifiable
    • Shareable - git-versioned, PR-reviewable
  • Vibe coding is still fine for some parts.
  • Separate the design from the build of a project.
  • Use AI to generate the code that fulfills your documented specifications.
  • A spec formulates your intent.
  • Updating your specs triggers the regeneration of code.

Talk: What 500+ Production Environments Taught Us About Shipping AI Agents

by Liran Hason

  • Treat every model upgrade like a migration - not just as changing a string.
  • A new model is a new teammate.
  • Invoices will arrive.
  • Check cache ratio not only number of tokens.
  • UX matters - show to the user what the agent is doing - this is the line between users complaining about slowness and stop getting complaints.

Talk: Owning the Inference Layer: When and How to Run your Own Models

by Taylor Smith

  • Chossing the right model and hardware setup:
    • Model is hoted by a provider. Charged absed on usage e.g. GTP, Claude, Gemini
    • Close commit. e.g. Google Vertex AI, Azure AI Foundry, Amazon SageMaker
    • Host your own model. You manage the scaling, serving, and hardware utilization.

My recap

  • Thanks to all the speakers for sharing your experiences and knowledge.

  • Thanks to WeAreDevelopers for such a great organisation and leading your values by example.

  • My recorded presentation at WeAreDevelopers World Congress 2026 Europe - Virtual Stage AI Agents & Agentic AI