The Work We Want: AI, Code Directors, and Collective Power
Collection :
A keynote about AI that was not about AI
At NEDCamp on November 14, 2026, Ethan Marcotte gave a keynote called "The Work We Want." Ethan coined the term "responsive web design," so a room full of Drupal people might have expected a talk about layouts. Instead, he told us up front that this was a talk about AI that was not really about AI. It was about our relationship to our work, and above all about power and hope.
I have spent much of this year writing about how AI is changing the work of Drupal developers. Ethan is openly skeptical of the AI-first future our industry keeps announcing. He might not agree with the conclusions I have drawn from my explorations of working with AI, but the questions he raised are important, particularly as the rate of change continues to accelerate.
From liberation to efficiency
Ethan built his talk on The Real World of Technology, a book by the physicist and social critic Ursula Franklin. In his paraphrase, every technology passes through three stages: advocacy, adoption, and institutionalization. First it promises to free us from drudgery. Then standards and infrastructure grow up around it. Finally it becomes an expectation, controlled by a small number of powerful players, and the original promise of liberation goes unfulfilled.
His example was the sewing machine. In 1860, one writer predicted it would banish ragged clothing from every class of society. A few decades later, manuals were advertising 3,300 stitches per minute. The machine found its real home on the factory floor, where garment work was split into narrow tasks, and each task was eventually automated. "People are always promised that technology will free them," Ethan said. As the technology matures, it captures them instead.
He argued that AI is following the same arc, only faster. He walked through companies that made AI use a baseline expectation, tied new headcount to what models could not do, or scored employees on their attitude toward AI. Then he walked through the reversals: leaders admitting that much of the output was low-value "slop," and one company rehiring hundreds of human support staff two years after cutting those roles in anticipation of AI gains.
Whose benefits, whose risks?
During the Q&A, someone asked how to balance a technology's potential for misuse with the opportunity it creates. Ethan answered with another idea from Franklin. Instead of asking about the benefits and risks of a technology, ask whose benefits, and whose risks.
That reframing stuck with me, because it resonates with an idea I’ve been exploring. In Part 3 of my Vibe Coding Drupal series, I introduced the Code Director: a developer who provides direction to an AI agent, or even a swarm of agents, the way an Art Director guides designers toward a vision. In The Pragmatic Middle Path, I described three ways developers respond to AI. AI rejection says, "I will do it myself." Vibe coding says, "I will offload it." The Code Director says, "I will delegate it," and stays accountable for every line that ships.
Seen through Ethan's lens, offloading is the factory model. Break the work into tasks, automate each one, and treat the person as a cost to remove. It looks fast, but it quietly piles up cognitive debt and intent debt, and the result is the kind of slop his examples described.
Code Direction is a different arrangement. The developer's judgment becomes the most valuable part of the work: setting context, scoping the task, watching the agent, testing the result, understanding the code, and validating it against Drupal's standards. The time saved goes to the person doing the work, and so does the responsibility. It feels much more like the design studios where I worked as an Art Director than the factory floor Ethan described.
That argument has limits. A job title does not decide whether AI frees people or captures them. That depends on choices made by the people who build the tools and the organizations that adopt them, which is exactly Ethan's point.
Directing agents is not just for developers
If Code Direction is a better answer to "whose benefits," it should not stop at the developer's desk. Marketing teams face the same pressure to do more with AI, and the same risk of being treated as the cost to remove.
Acquia Source is the open digital command center that connects content, digital assets, and governance workflows with AI and the martech stack teams already use. Teams reach it through Acquia Source CMS, Acquia Cloud Platform, or Acquia DAM. Inside it, Acquia AI provides agents that take on marketing operations work, under the same permissions and approvals that already govern the people on the team. Here is what that looks like today:
- Acquia AI Advisor is a knowledge agent, available now, that answers questions about Acquia products, Drupal, and support from Acquia documentation, so campaign teams get answers without filing a ticket.
- On Source CMS, an agent can take a brief, read your brand guidelines without being told to, build a page from your site's existing components, and save it as a draft. This is in Limited Availability.
- With Acquia Web Governance, an agent checks pages for accessibility against WCAG, data privacy, readability, and brand policy, and can act on what it finds on a Source CMS site. This is in Limited Availability.
- With Acquia DAM, an agent can search the asset library and recommend approved assets that fit the content. This is in Limited Availability for new DAM customers.
The marketer's job in this picture mirrors how I think about the Code Director. You supply the brief and the context, watch the work, review it, and decide what goes live. A person approves anything a visitor would see. Each agent signs in under its own identity, reaches only what an administrator assigns, and shows its activity for review before approval. The agent handles the drafting and checking; your team still decides what ships.
Reasoning that improves week by week
The reasoning ability behind these tools keeps improving at a pace that seems to show up week by week. Tasks that needed careful hand-holding when I started this series now come back right the first time more often than not. Agents plan further ahead and catch more of their own mistakes.
As I wrote in AI as a Reasoning Partner, if your only contribution is the prompt, your value will fade as models improve. Judgment holds its value. However good the agent gets, someone still has to know what to ask for, tell whether the result is right, and own the decision to ship it.
It also means nobody should have to spend their week chasing model releases. Acquia selects and manages the model behind Acquia AI, which is what lets the governance, data, and audit guarantees hold. Marketers and developers can keep their attention on direction and review.
Hope is participatory
Ethan closed with collective action. He described tech workers who identified something they wanted to change at work and organized to change it, through open letters, shareholder resolutions, walkouts, and union contracts that now cover everything from pay to AI policy. He also shared a line from Cornel West that has stayed with me: hope is not the same thing as optimism. Optimism waits for the evidence. Hope is participatory.
Drupal people already know a version of this, because it is how our software gets made. Thousands of contributors decide together what Drupal becomes, through issue queues and community initiatives. It’s how we have worked together to build Drupal 12, which will launch in the coming weeks. That same culture is shaping how our community uses AI. The AI Best Practices for Drupal project gathers our collective expertise into skills that ground AI coding agents in Drupal conventions. Our coding standards, static analysis, and automated tests give Code Directors the guardrails to verify what agents produce. We are writing the norms for AI-assisted Drupal work ourselves, in the open, instead of waiting for someone else to set them. There have been important and impassioned conversations this year about the role AI should have in Drupal’s future. The fact that this question is being answered collectively, instead of by one company, will be an important cue in determining who benefits.
Ethan ended by swapping one question for another. Instead of asking what the future will look like, ask what we want it to look like, and then answer that question together. For developers, I hope the answer looks like approaching this new paradigm for development as Code Directors. For marketers, I hope it looks like a team that directs its agents rather than being replaced by them. If you were at NEDCamp, or you are working through these questions with your own team, I would love to hear how you answer it.