For interactive AI, that changes where design happens. We’re no longer binding digital interactions within predetermined visual interfaces. Many of the decisions that shape an AI experience get made at runtime, by the system, in situations nobody defined in advance. You can’t script all possibilities any longer. What you can do is decide what the system should favor when it has to trade one option for another.
Designing is deciding. Deciding is taking risks about the future. Success depends on acting accordingly. Trust depends on getting it right.
Lack of trust is a major issue for technology, and it matters more the more impact a system has and the less understandable it is. We might get frustrated at the feeds our social media platforms give us. That’s less significant than trying to communicate with a financial institution through AI. When health or money is at stake, we expect transparency, predictability and reliability.
Principles are a go-to response from many design and product teams. Be clear, be accurate, be fast. They describe what good looks like, often generically, and there’s some usefulness in that. But they’re toothless, and they leave out the hard part. In a real decision, nobody is arguing about whether clarity is good. The question is how much simplicity you’ll give up to get it. Or how much low latency costs.
Tenets answer the question: how do we deliver what makes sense, and deliver it well? A tenet is about managing the tension between elements of a system. Characteristics, outcomes, resources. Each tenet names two things that pull at each other and says which one gives way when you can’t have both to the degree you want. They are often uncomfortable to write out. “Over speed” is a commitment to being slower than someone wants sometimes. Managing the discomfort is design. It makes the cost explicit, so a team can negotiate it before a user pays it.
Tenets also force precision of meaning and impact. Speed shows up twice in the list below. “Predictability and negotiation over speed and ease” puts it second. “Response speed over accuracy” puts it first. Both can hold only if they mean different kinds of speed, and working out which kind is which is the useful part. Naturalness shows up on both sides too. Each tenet leans on working through the tension, and where the decisions land reflects the values of teams and companies.
With AI, tenets can do a second job. Beyond guiding the team, they can be written into the agent’s instructions, as the decision principles it uses when it meets a situation a human didn’t explicitly plan for.
Tenets also drive what you measure. If you choose clarity over simplicity, you should be able to see the complexity you took on and check that it paid off. A tenet that never costs you anything visible isn’t doing any work.
These are the tenets I wrote for conversational AI, and use frequently in my work:
- Predictability and negotiation over speed and ease
- Clarity over simplicity
- Coherence and cohesion over surface consistency
- Response speed over accuracy
- Comfort and brevity over density
- Transparency over artificial naturalness and emotion
- Meaning and naturalness over custom and perceived rules
- Relevance and benefit over branding and sales
- Explainability over power and technological advancement
- Simplicity over efficiency
- Accuracy over precision
- Naturalness and flexibility over deadlines
- Impact over volume
- Respect over expedience
If you’re deciding how an AI system should behave, here’s how to write your own:
- Start from the decisions your team keeps arguing about. A recurring argument is usually a tension nobody has resolved.
- Name both sides as things you actually value. A tenet only works if what gives way is also good.
- Choose what matters for your context, and note what’s at stake that makes you lean that way. Money and health matter differently than entertainment.
- Test each one against a real decision you’ve already made. If the tenet wouldn’t have gotten you to a better decision faster, it might still be just a principle.
- Put them into use where decisions get made: reviews, requirements, critiques, and the agent instructions.
- Measure the cost you chose to pay, and revisit a tenet when it stops paying off.
Over the next few posts I’ll take two or three at a time, with examples of the tension in real decisions.
Is your product performance stuck in team arguments that never lead anywhere?