Notions

There is a particular kind of knowledge that lives in the hands. Watch a concept artist block out a scene, build a procedural setup for the way a plant grows, or add a few quick details to a sketch mid-meeting so a room finally sees the same direction. Behind each is years of accumulated craft, specific to each tool and each problem, hard-won and largely invisible. I used to think that was the job. Increasingly, I think it was only the venue. The trajectory of agentic AI - and, before long, of intelligence built into the tools themselves - points towards a near future in which the software works itself, on our behalf, reaching past the screen into the world. An agent navigates the panels, wrangles version history, contacts a supplier or helps realize an idea - all the while we attend to intent. Human-computer interaction, as we have known it through the user interface, begins to dissolve - redrawn for two audiences at once: the person, and the agent. For anyone who spent years learning task procedures, workflows, and the private liturgies of particular tools, this is a bittersweet shift - all that fluency, quietly automated away. But it may also be the dawn of the generalist: an era of decentralized, cross-disciplinary design, where the walls between the specialisations come down. On that note, there is one consequence that deserves more scrutiny than it usually receives. Much of it was judgment: why this camera angle works, what temperature a mould requires, which operation must precede another. Often, we learned to make those decisions through the work itself. Repetition trained perception as much as dexterity. The concern, then, is what happens when automation absorbs the judgment along with the procedure. If a system sets the mould’s temperature without making its reasoning available, the practitioner may be left approving a result they no longer know how to assess. And if learning the operation once helped develop that understanding, how will the next practitioner acquire it? Breadth without comprehension can become dependence. A renaissance requires more than access to the tools. This changes what we need to design. When agents operate software on people’s behalf, the interface becomes a place where authority is negotiated between the person, the agent, and the system. What may the agent decide alone? When must it hand a decision back? What should it reveal about its assumptions, uncertainty, and consequences? These are design problems of legibility, delegation, and accountability. The person responsible for an outcome needs ways to understand and challenge the decisions that produced it. So what should a practitioner deepen as more of the operating becomes automated? The earlier posts offered the first half of my answer: direction, taste, and vision, guided by moral judgment. The second half concerns how we put those qualities to work. Learn to brief precisely: communicate what matters, what can change, and what a successful outcome would mean. Learn to evaluate carefully. When the tools produce a hundred variations before lunch, choosing deserves more attention, not less. And learn to interrogate what they hand you—to recognize when an answer is plausible, when it is sound, and when you lack the knowledge to tell. In my own work, pairing Houdini with generative tools to build worlds and moving images in service of story, I can explore variations, atmospheres, and visual set-ups that would once have exceeded the time I had available. I can stay with an idea longer before committing. I can try the version I would otherwise have left unexplored. That possibility matters to me. It makes a broader practice imaginable, with more room to experiment across disciplines. But its value still depends on the understanding each designer brings to the table, and on what we are willing to learn along the way. The tools extend our reach; experience helps us judge where to take it. There is something of a renaissance in that possibility, provided we remain curious about how things work. The knowledge in the hands was never incidental. It helped form the eye, the judgment, and sometimes the idea itself. As we delegate more of the execution, we need to find ways to keep that understanding alive. Spend the hours we gain on direction, on inquiry, on the version we have not yet tried. Keep asking what the machine is deciding and whether we understand enough to stand behind it. Then push further. Building something worthwhile will still depend on what we understand, what we question, and what we ultimately choose to make.