Meta description: A profile of Delon Swartz on AI product design practice — agent UX, shipping discipline, and building AI-native products from South Africa without the hype cycle.
AI product design is not “UI for chatbots.” It is the discipline of turning probabilistic systems into reliable products: clearer intent, safer actions, measurable outcomes, and interfaces that respect human attention.
Delon Swartz is an AI Design Engineer and product builder in Durban, South Africa. He works across Applied AI, Applied ML, and AI product design; founded ModusMax AI; and ships end-to-end products through Delon Labs UX. This essay shares his practice — how he thinks about agents, UX, and delivery — without fabricating portfolios or vanity awards.
The object of design has changed
Classical digital product design assumed deterministic software. Click → known state. AI-native products assume:
- Outputs vary.
- Tools can take actions with side effects.
- Context windows and memory change behaviour over time.
- Users must negotiate control with the system.
That means AI product designers (and hybrid AI design engineers) inherit concerns that used to sit only with SRE, security, and ML ops: permissions, observability, evaluation, and graceful degradation.
If a design system has a button component but no pattern for “agent is about to modify a file — confirm?”, it is not yet an AI design system.
Practice pillar 1: Intent over prompt theatre
Users should not have to become prompt engineers to complete a job. Good AI product design:
- Offers structured intents (“Summarise this thread for legal review”) alongside free text.
- Captures constraints in UI (tone, length, data sources, deadline).
- Turns successful prompts into reusable skills or playbooks.
In agent platforms, this shows up as skills, spaces, and clear task framing rather than a blank box and hope. ModusMax’s product direction — desktop-first agents, local or cloud engines, chat plus operational surfaces — is built around that belief: power without terminal gatekeeping.
Practice pillar 2: Agency with brakes
“Agent” means the system can plan and act. Product design must decide how much agency is unlocked at each trust level.
A practical ladder:
- Suggest — draft only.
- Assist — act inside a sandbox.
- Execute with confirm — propose a tool call; human approves.
- Execute within policy — autonomy inside an allowlist.
- Supervise — human reviews exceptions and audits.
Climbing this ladder too fast is how demos become incidents. Climbing too slowly is how products feel useless. The craft is matching ladder height to domain risk — a theme that runs through Swartz’s agent UX work.
Practice pillar 3: Make the machine’s mind legible (enough)
Full transparency of every token is rarely usable. Strategic legibility is:
- What sources were used?
- What tools were called?
- What is the plan before irreversible steps?
- How does the user stop, undo, or take over?
Logs and timelines are UX. So are empty states that teach mental models (“Can read these folders; cannot send email until Gmail is connected”).
Practice pillar 4: Design the boring path
AI product marketing loves the wow path. Real users live on the boring path: login, permissions, partial failures, unknown/uncertain answers, rate limits, offline mode, and handoff to a human.
Swartz treats the boring path as first-class:
- Explicit failure copy.
- Partial success states.
- Export and escape hatches.
- Support surfaces that do not require the user to screenshot a raw stack trace.
This mindset comes as much from shipping full-stack products (UX through backend and deploy) as from model work. AI features that cannot be operated will not be loved.
Practice pillar 5: Build with AI, design with judgement
Modern builders use AI-assisted development tools. Swartz does too — they compress the distance from concept to working software. That speed is valuable. It is also dangerous if judgement does not keep up.
AI-assisted coding does not absolve product ethics, accessibility, security reviews, or taste. His practice is: move faster on implementation, stay slower on claims. Especially in public writing, he prefers precise statements over inflated ones. Markets punish fiction.
South African vantage point
Designing AI products from South Africa is not a footnote. It is a vantage point that shapes his work:
- Global tooling, local users — ship with world-class models and frameworks while designing for African bandwidth, pricing sensitivity, and mobile habits where relevant.
- Founder-scale ownership — many SA builders wear design, ML integration, and delivery hats in one role. That hybrid is exactly where AI Design Engineering lives.
- Narrative responsibility — the market does not need imported hype cycles. It needs working agents, honest UX, and products like job platforms that serve real people (for example, youth-focused tools such as TheJobPlug).
He has also shared public concept work — from streaming co-viewing UX explorations to notes on video agents — because product practice improves when ideas meet critique early.
A weekly rhythm that keeps practice honest
For individuals and small teams, a lightweight rhythm beats a heavy process cult:
| Cadence | Focus |
|---|---|
| Daily | Ship a thin vertical or kill a bad assumption |
| Weekly | Review eval samples + 3 real user/session notes |
| Bi-weekly | Permission and safety review for new agent tools |
| Monthly | Re-read positioning: are we solving a job or chasing a demo? |
Write in public when possible. A clear LinkedIn article or Dev.to post forces sharper thinking than an internal wiki nobody reads — a habit Swartz models in his own posts.
What he is building toward
ModusMax is his clearest statement of AI product design belief: agents should be operable by people who do not live in a terminal, with local and cloud options, and product surfaces that make agency understandable. Early access details are on modusmax.online.
Alongside that, Delon Labs UX continues the habit of designing and shipping complete products — because AI product design that never touches production is only theatre.
Connect
For product, design, or Applied AI leaders seeking a practitioner conversation — South Africa–based, globally minded:
AI product design will keep changing with every model release. The durable skills are older: empathy for users, respect for risk, and the discipline to ship.
About
Delon Swartz is an AI Design Engineer in Durban, South Africa, focused on Applied AI, Applied ML, and AI product design. He is the founder of ModusMax AI and builds full-lifecycle products via Delon Labs UX. Connect on LinkedIn.
Top comments (0)