The Life of a Digital Agency in the Time of AI - Part 2
This is the second post in a two-part series on the life of our agency in the time of AI. In this part, we explore the human element behind every project. Read Part 1 here.
The Work that Takes Real Time, Real Effort, and Real People
In Part 1, we looked at the latest wave of hype surrounding generative AI — and how it mirrors earlier cycles of inflated expectations around "free" websites and "magic" cloud computing.
Here in Part 2, we'll shift focus to something more grounded: the work that still requires people. Not just because the tools aren’t there yet — but because what’s required is nuance, context, conversation, coordination, and care. These are the things that actually go into a successful digital project — whether or not AI tools are in the mix.
We've also previously written about the importance of Expectations and Engagement .
To be clear, AI-based tools are coming, and they're going to redraw the software development landscape, but how quickly that happens, and with what 'bumps or crashes' along the way is the subject of heated debate.
1. Project Mission, Vision, and Scope
Every successful project starts with a conversation — and that conversation can’t be replaced by a prompt. It’s where we uncover why the project matters, who it’s for, what it should achieve, and what constraints we’re working with.
Sometimes this phase is about helping the client clarify their own thinking. Other times it’s about translating ambition into something focused and feasible — often with an eye on future phases of work.
It’s about drawing out the value proposition, negotiating tradeoffs, identifying risk, and agreeing on what done looks like. These are strategic conversations, not outputs.
Could this be done with AI? I think the "echoed output" in an AI-based conversation about all of the above might help. It might add dimension to the discussion, perhaps even overlooked points of view. But we're pretty sure that for the moment - it requires humans to synthesize the results and ultimately decide on the best way forward.
2. Project Organization
Once goals are set, the next challenge is structure — not just of the product, but of the working relationship.
Clients often underestimate the cost of coordination. It takes real effort to communicate progress, surface issues, test assumptions, and keep momentum going. Modern collaboration tools help enormously — but they don’t replace the need for actual engagement.
Meetings, updates, emails, messages — all part of the invisible scaffolding (and real costs) that holds a project together. Getting this right saves time. Getting it wrong derails projects. Again — AI can help, but it can't replace the need to organize, steer and validate results against stated objectives.
No tool can automate alignment.
3. Design
AI can help brainstorm ideas. It can riff on a visual theme, mock up a layout, or generate icons and illustrations.
But we would argue that design is about intent — aligning visual direction with brand values, user needs, and strategic goals. That means research: mood boards, pitch sketches, competitive audits. It means feedback: iteration, critique, refinement. And it means output: assets in every size, shape, and format for the web, for mobile, for social.
As unexciting as this may sound - creating correctly sized open graph images, banners, app icons, browser favicons and other production-ready assets that all look correct in multiple channels takes time. Not just to make, but to make right.
We've not seen an AI tool yet that can correctly generate assembly-ready assets, in either final output, or in an output that demonstrates continuity between generations.
AI can help, but it won’t yet give you a cohesive, intentional, polished system that holds up across screens, channels, or other outputs.
4. Development and Assembly
Development isn’t just about translating designs into code — it’s about interpreting them wisely. That means making good decisions about architecture, components, and structure. It means choosing storage formats, frameworks, libraries, platforms. It means knowing when to reach for a tool — and when not to.
Code can be generated. But architecture is planned. Clean, readable, well-organized code — built with future enhancements in mind , comes from people with a deep understanding of solutions architecture, and who care deeply about quality and craft. Tradeoffs are weighed, and only experience creates robust and maintainable solutions.
Even if customers shouldn't need to care about the 'craft' - they are certainly going to care when their system breaks (at best), or is left vulnerable to a critical security issue resulting in the loss of PII (at worst).
5. Deployment
Every deployment decision involves risk. Where will the system run? Who’s the audience? How many users are expected — and what happens if it’s more than that? What’s the plan for security, rate limiting, content delivery, caching, backup, and failover?
Good deployment isn't just about pressing “Deploy.” It’s about knowing how to design infrastructure for cost, reliability, and scale. It’s about applying best practices in overall security posture as well as understanding legal obligations.
It’s about protecting clients not just from downtime — but from surprise bills, attack surfaces, and future regrets.
This Still Isn’t Push-Button Work
The common denominator in all of the above? Nearly every phase of a digital project involves conversation, alignment, expertise, and judgment. Yes, AI can assist — and we use it where it helps. But, as we described in Part 1 of this two-part post, it takes experience to know when AI is getting it right, or wrong.
That’s what clients hire us for. That’s what we show up for. And that's why it's more important than ever to remind ourselves — and our clients — what real digital work still looks like.
It looks like people thinking.
It looks like teams working.
It looks like care, applied deliberately, over time.
And for now — and likely for a while to come — there’s no AI shortcut for that.
Read Part 1 of this this two-part post.