When are you getting your Agent ?
People don’t want more apps; they want outcomes. The next era of the consumer internet shifts the primary interface from tapping through screens to stating goals—“rebook my trip,” “return this,” “lower my bill”—and having an agent get it done. In this post-app world, consumer experiences are orchestrated by AI that can plan, negotiate, transact, and confirm across services—without app-hopping or forms.
What changes isn’t just UI. It’s architecture, discovery, and trust:
From pages to capabilities. Providers expose what they can do (“rebook,” “refund,” “replace”) rather than just pages to visit.
From sessions to outcomes. Value is measured by resolutions, not clicks or time-on-site.
From UI chrome to trust primitives. Permissions, audit logs, and human-in-the-loop controls become the core UX.
From search to intent routing. Discovery moves to capability registries and agent-to-agent protocols; execution becomes dual-mode: API when available, GUI automation when not.
The consumer-agent pipeline (in one loop)
Intent capture → parse and disambiguate what the user wants.
Planning → select tools/providers, fetch policies, price options, pick paths.
Execution → transact via API or reliable GUI automation, escalate when needed.
Confirmation → deliver receipts, savings, state changes, and a full audit trail.
Learning → improve prompts, preferences, provider choices, and guardrails.
This loop repeats, compounding into a household agent that “just handles it.”
Five areas to watch (and why they matter)
Shopping moves inside agents
Checkout is shifting into chat-style agent UIs. The interface that explains, compares, and answers questions is the same interface that pays. Why it matters: Lower drop-off, richer context at purchase, and new attribution models tied to outcomes.Phones become agent-first
OS-level assistants and on-device models are being wired into Shortcuts/intents so “do this for me” routes across apps automatically. Why it matters: The operating system becomes the router for consumer intent—not the app grid.Homes go agentic
Household assistants are learning to orchestrate across devices and services—and even complete web flows when APIs don’t exist. Why it matters: The home is no longer a bundle of gadgets; it’s a unified execution surface.Distribution piggybacks on fintech
High-frequency payment apps are experimenting with agentic search and assistance. Why it matters: Agents gain daily consumer surface area (pay, split, subscribe), collapsing discovery and transaction into one step.Platforms reframe the web
“Agentic web” efforts (NL-style interfaces, capability protocols, multi-agent orchestration) position agents as the primary UI and the web as a network of actions. Why it matters: Standards, not scrapers, will decide who gets picked for execution.
UX for trust (the new product surface)
Granular permissions: per-provider, per-action, expiring by default.
Transparent plans: show the proposed steps before acting; allow edits.
Receipts and reversibility: itemized logs, one-tap undo, and dispute flows.
Escalation moments: clean handoff from agent to human (and back).
Data minimization: fetch the least needed to complete the job.
Business model shifts (beyond ads and installs)
Outcome-based economics: fees on verified resolutions (refund issued, rebooking completed, savings captured).
Capability registries: a new “SEO” where providers publish machine-readable actions and service levels.
Agent SLAs: reliability, latency, and compliance become differentiators.
Brand presence in agents: product cards evolve into capability cards—clear terms, prices, and guarantees.
Risks and open questions
Lock-in vs. openness: Will a few platforms control capability routing, or will protocols keep the playing field open?
Consent fatigue: How do we design permissions that are safe and sane?
Hallucinated actions: What are the right guardrails before money moves?
Attribution and accountability: When things go wrong, who owns the outcome—the agent, the platform, or the provider?
How to prepare (whether you’re a brand or a builder)
Expose capabilities, not just pages. Publish what your service can do in a machine-readable way.
Design for agent clients. Clear policies, sandboxed test flows, fallback paths when APIs are incomplete.
Ship trust primitives. Permissions, scopes, audit endpoints, and fast dispute resolution.
Instrument outcomes. Track resolutions, not sessions; measure savings, time restored, and NPS change.
The bottom line
The consumer internet is pivoting from navigation to negotiation, from clicking to getting things done. Agents won’t replace every app overnight—but they’ll become the default entry point for high-friction, multi-step tasks. The winners will be those who treat trust as interface, capabilities as content, and outcomes as the metric that matters.
Events
I am speaking at https://www.imaginationinaction.co