Big Ideas 2026: The Agentic Interface

Marc Andrusko argues the prompt box is dying as the primary AI interface. The next wave of AI apps will proactively observe user behavior and suggest actions for approval, similar to high-agency employees who identify problems, research solutions, and present recommendations. This shift expands the addressable market from $300-400B in software spend to $13T in labor spend.

• •AI apps will move from reactive prompting to proactive intervention - observing work and suggesting actions like S-tier employees who solve problems independently
• •Market opportunity expands 30x from traditional software ($300-400B) to labor replacement ($13T in US alone)
• •Future AI CRMs will autonomously identify opportunities, harvest old leads from emails, draft communications, and manage pipelines - requiring only final human approval
• •Power users will train AI with larger context windows and memory to achieve 99.9% autonomous task completion without human approval
• •Human-in-the-loop approval will remain standard for high-stakes contexts, but models are capable of reaching the 'click accept' threshold

Stephanie Zhang introduces the concept of machine legibility as the new design paradigm. As agents become intermediaries between users and systems, visual hierarchy becomes less important than structural clarity. Content creation, application design, and information architecture must optimize for agent comprehension rather than human attention patterns.

• •Visual design and UI optimization matter less when agents navigate software - the focus shifts from human eyes/clicks to machine-parseable structure
• •Traditional content strategies (hooks, lead paragraphs, search ranking) become obsolete - agents won't miss relevant information buried on page five
• •AI SREs now analyze telemetry data and report insights directly to Slack instead of requiring engineers to navigate Grafana dashboards
• •GEO (Generative Engine Optimization) tools are emerging to help organizations appear in ChatGPT responses and agent-driven searches
• •Risk of low-quality, high-volume content spam as creation costs approach zero - similar to keyword stuffing but for agent attention

Discussion of where human oversight remains critical versus where agents can operate fully autonomously. Customer support (Decagon) already runs autonomously, while security operations and incident resolution maintain human-in-the-loop due to higher liability and complexity requirements.

• •Decagon already answers customer questions autonomously without human intervention in production
• •High-liability domains (security ops, incident resolution) keep humans in the loop - agents analyze and present options, humans decide
• •Human involvement will persist longer in complex analysis scenarios until models reach extremely high accuracy thresholds
• •Content optimization shifts from capturing attention with flashy hooks to maximizing insight and relevance for agent consumption

Sarah Wang describes how a dynamic agent layer is emerging above traditional systems of record, fundamentally threatening their data gravity advantage. Using IT service management as a case study, she explains how collapsing the distance between intent and execution creates 10x better experiences, not just 20-50% improvements.

• •First genuine threat to systems of record dominance - previous SaaS 2.0 attempts failed by only offering better UI (20-50% improvement)
• •ITSM example: IT head with 20-year career believes support will fundamentally change in 5 years due to agent-based systems
• •New ITSM agents make software access requests nearly instantaneous by extracting intent, classifying requests, mapping workflows, and identifying entities
• •Agent layer sitting closest to users and collecting preference data will accrue the most value - not the foundation models or legacy systems
• •Startups winning against incumbents (AI SRE companies like Resolve and Traversal beating Datadog-based agents) because they can iterate weekly/daily
• •Success requires accurate, reliable solutions to build user trust - speed of iteration is critical competitive advantage

Synthesis of the three big ideas into a unified vision: interfaces shift from chat to action, design shifts from human-first to agent-readable, and workflows shift from systems of record to agent layers. AI transitions from a consultative tool to an autonomous execution system that turns intent into outcomes.

• •Interface evolution: from prompt boxes to proactive action suggestions that users approve
• •Design evolution: from visual hierarchy optimization to machine-legible structure and data
• •Workflow evolution: from passive systems of record to dynamic agent layers that execute on user intent
• •Core transformation: AI stops being 'something you ask' and becomes 'something that does' - autonomous execution vs. consultation