But We’re Still Accountable.

The Work Has Changed.

So much more is possible today, but it still takes a human in the loop to ask the right questions and stand behind the results.

The Work Has Changed

Now you can make almost anything, almost instantly.

What used to take us weeks takes minutes. Production work is machine work. Code, copy, imagery, whole animated interfaces (like this one), can be generated on demand. The execution economy has gone bankrupt. The value is no longer in the doing. It’s in the thinking.
The sky’s the limit. The possibilities are endless.

Every idea is a natural language conversation away from entering reality. Every prototype has the potential to become a product. There is a great power at our fingertips. More is possible than ever before.
With more possibility comes more risk.

The more moving parts we deploy, the more opportunities for them to break. No one is infallible—even the machines (at least for now). When anything can be built, deciding what to build, and why, becomes just as important.
Prompting blindly creates a tangled mess.

Vibe coding can only get you so far. Complexity accumulates, and with nobody understanding what’s actually under the hood, it’s only a matter of time before the system collapses under strain.
That’s why you need a human in the loop.

Responsible design encodes intent, constraints, and architectural guidelines into the agentic model. We know what agents are doing, and what they’re building. AI accelerates our work. But experience steers it.
Don’t lose sight of the big picture.

Quick answers have always been a commodity. But the big picture is an expensive one to ignore.

I spent a long time in the weeds, writing code by hand, grappling with edge cases, and designing architecture from concept to execution. That experience helps me evaluate AI-generated implementations for feasibility, stability, and longevity. Agents can’t hold all that context in memory. But humans still can.

A Process for the Future

An iterative process,
for the agentic era.

Six connected practices, revisited as we learn: a loop of ideation. At the end of that process, we start all over. Launch is a milestone. Here’s how it breaks down.

Intuit

Intuit the problems, explore the solutions. We leverage human judgment continuously to drive the discovery process early and often. AI acceleration helps us discover hidden insights and double back and test without losing momentum.

Then: Discovery as theory, and without practice.

Weeks of stakeholder interviews that produce a requirements doc nobody reads. We arrive at the “solution” before anybody tests for the problem, or understands it.

Weeks of stakeholder interviews that produce a requirements doc nobody reads. We arrive at the "solution" before anybody tests for the problem, or understands it.
Now: Intuit the problems, explore the solutions.

AI helps us sift through mountains of research, analytics, and noise in hours instead of weeks. We intuit which signals matter and present concepts worth testing. Discovery becomes a continual process.

AI helps us sift through mountains of research, analytics, and noise in hours instead of weeks. We intuit which signals matter and present concepts worth testing. Discovery becomes a continual process.

Design

Transform concepts into tangible blueprints. Design encompasses everything from visual and UX design to technical architectural plans.

Then: A relay race of artifacts that lose design intent as time passes.

Static sitemaps, wireframes, design comps, and other visual artifacts, moved down the pipeline one handoff at a time, losing design intent with each handoff, like a game of telephone.

Static sitemaps, wireframes, design comps, and other visual artifacts, moved down the pipeline one handoff at a time, losing design intent with each handoff, like a game of telephone.
Now: Design the whole system at once.

Information architecture, high-fidelity visual design mockups, and technical blueprints evolve together as working prototypes that we can put in front of people early. GenAI makes creating options cheap. Taste and research drive decision-making.

Information architecture, high-fidelity visual design mockups, and technical blueprints evolve together as working prototypes that we can put in front of people early. GenAI makes creating options cheap. Taste and research drive decision-making.

Execute

Put human domain experts and bounded agents to work against a shared blueprint. Human judgment drives acceleration.

Then: Everything built by hand, one agonizing sprint at a time.

Months of hand-writing code, bottlenecked by human headcount. By the time the build catches up and priorities change, we can’t adapt without it costing an arm and a leg.

Months of hand-writing code, bottlenecked by human headcount. By the time the build catches up and priorities change, we can't adapt without it costing an arm and a leg.
Now: Agentic development accelerates our work, making more possible.

Humans spend time architecting while agents spend time implementing. Each agent takes a bounded assignment under strict guidelines, and the self-documenting environment grows while remaining comprehensible.

Humans spend time architecting while agents spend time implementing. Each agent takes a bounded assignment under strict guidelines, and the self-documenting environment grows while remaining comprehensible.

Automate

Automate the work that endures. Never waste human labor on work machines can do for us. Our time is valuable, and our judgment priceless.

Then: We took on the burden of mindless tasks because there was no better way.

People copied, pasted, reformatted, and re-reviewed by hand, slowly, and expensively. Human error was inevitable where tedium reigned.

People copied, pasted, reformatted, and re-reviewed by hand, slowly, and expensively. Human error was inevitable where tedium reigned.
Now: The machines have our backs, so we can focus on work that matters.

We find the work AI can handle (mass migration, data handling and transformation, integrity verification, complex but predictable workflows) and connect it to the swarm. Agents report back when they get stuck, and we spend less time babysitting processes.

We find the work AI can handle (mass migration, data handling and transformation, integrity verification, complex but predictable workflows) and connect it to the swarm. Agents report back when they get stuck, and we spend less time babysitting processes.

Test

Challenge our assumptions, and those of our agents. Always seek evidence. Test quality, test security, test accessibility, test reliability. Testing is a way of life.

Then: Neglected QA creates crunch time at launch.

Execution was already rushed, and quality testing cut short whenever the schedule slipped. The product’s first contact with real visitors might be right after it shipped, rather than during UAT.

Execution was already rushed, and quality testing cut short whenever the schedule slipped. The product's first contact with real visitors might be right after it shipped, rather than during UAT.
Now: Evidence at every step.

The agentic environment allows us to incorporate testing mechanisms that produce evidence of intended behavior as every new feature is developed. Rapid prototyping enables security and accessibility audits and advances insight into the application, which means real people see the work early. We trust the evidence, not assumptions.

The agentic environment allows us to incorporate testing mechanisms that produce evidence of intended behavior as every new feature is developed. Rapid prototyping enables security and accessibility audits and advances insight into the application, which means real people see the work early. We trust the evidence, not assumptions.

Evolve

The system is always evolving. We stay in conversation with it about what to improve and what comes next.

Then: Launch, then pencils down.

The site goes live and the team moves on, paying lip service to its continued welfare. The product slowly decays as software updates, bugs, and faulty workflows accumulate crud, until budget comes along for a redesign.

The site goes live and the team moves on, paying lip service to its continued welfare. The product slowly decays as software updates, bugs, and faulty workflows accumulate crud, until budget comes along for a redesign.
Now: Launch is only a milestone.

Websites are not Nintendo cartridges: they’re changing every day as the web itself changes. We study the system in use (via analytics, logged failures, and human feedback), and extract what it’s teaching us to feed that knowledge straight back into the agentic loop and our next Intuit[ion].

Websites are not Nintendo cartridges: they're changing every day as the web itself changes. We study the system in use (via analytics, logged failures, and human feedback), and extract what it's teaching us to feed that knowledge straight back into the agentic loop and our next Intuit[ion].
Next stop: How do we keep the loop going?

The Work Continues

The loop can’t run on its own forever. Things fall apart.

That’s why you need a human in the loop.

The Human in the Loop

What do you desire for the future of your business?

Every turn of the wheel starts and ends with a human decision about what matters next, and why. Our intuitions, taste, and expectations are driven by lifetimes of experience. Domain experts put that experience into practice, charging agents with the character of their work.
A machine can tell you how it works. But only humans can tell you why it matters.
Agents extend our reach, making more possible.

Architects, researchers, implementers, testers, auditors, documenters—they’re like an army of genius interns, eager to help. But every eager intern needs a bounded assignment, and a duty to return evidence of the work done right.

Intuit Which question is worth answering?
Design Which option is right for our brand?
Execute Does the system do what it claims to do?
Automate Where should work stay in human hands?
Test What does the evidence prove?
Evolve What did the work teach us?

The Brief Defines Intent

Judgment & Context

Agents can only remember so much. The brief is the artifact that contains the most important details of the project: human intent and the long-term plan.
Humans Connect the Dots

Direction & review

The humans in the loop carry that intent across multiple domains, remaining responsible for how all the pieces work together.
Agents Extend Capacity

Execution & evidence

Bounded assignments in the hands of eager agents get the job done, under careful supervision.

Many specialists, and a single point of contact.

One accountable lead and a decentralized team of independent specialists, driven by empathy, decades of intuition, and on-the-ground research.

Responsible by Design

The loop preserves responsibility, at every step.

The conversation between humans in the loop and agents makes executing handoffs, accumulating evidence, and conducting reviews a recurring phenomenon.
We’re not reinventing the wheel, we’re replacing it with a jetpack.

Moodboards, wireframes, sitemaps, technical specs: these aren’t fossils of an antiquated era, but they are interposable. Agents can rapidly iterate on them as artifacts, so we can focus on the thinking they represent.
Where’s the responsibility if no one knows what’s going on?

Systems grow and increase in complexity with every new feature. They were hard enough to manage before the agentic era; now they can become nearly opaque when agents take over. Someone has to be able to understand what’s going on and fix things without guessing.
Who understands your users better than you?

Great user experiences require empathy for users. Your users’ needs and your business goals require someone to experience them in the real world, and champion them in every discussion.
Taste isn’t prompted, it’s cultivated over time.

GenAI makes creating options cheap. We can spin up 100 design directions for the price of 2. But knowing which one is the right fit for your business takes discernment that agents arrive at by consensus, rather than experience.
Human in the driver’s seat. Agents ride shotgun.

We put domain experts at the wheel, and put the roadmap in those genius interns’ laps, to help us look out for road bumps. You get the benefit of a safe but scenic ride.

Document everything, often.
It’s just as important that briefs capture the intent behind the system as the “how” behind its functioning. Technical details evolve, but that intent can get lost. Give both people and agents a shared foundation for the work.
Continually test assumptions.
Separate execution from evaluation. Inspect the actual behavior, ask for evidence, and keep consequential decisions with an accountable human.
Design for lifecycles, not releases.
Establish ownership, maintenance, and a way to respond when something changes. Continue to improve the system after it makes contact with real users.

A little human intervention goes a long way.

Complex systems, like cities, have a habit of producing situations nobody planned for. When the unexpected happens, human ingenuity makes all the difference.

Let's Get a Human in Your Loop

Get a free video consultation.

So much more is possible today.

One accountable lead and a decentralized team of independent specialists, driven by empathy, decades of intuition, and on-the-ground research.

Book Appointment

Consulting

The Work Has Changed.