AI Agents in 2026 | 3 Predictions For What’s To Come (a16z Big Ideas)
This episode explores three major shifts shaping the future of AI products. The discussion moves from the 'death of the prompt box' towards proactive AI that acts like a top-tier employee, to a new design paradigm of 'machine legibility' where we create for agents instead of humans. Finally, it covers the practical, real-world deployment of AI voice agents in enterprise sectors like healthcare and finance, signaling a move from AI as something you ask to something that does.
The fundamental nature of software is changing as AI transitions from a passive tool to a proactive agent. This evolution is marked by three key shifts: from prompting to execution, from human-first design to machine legibility, and from conversational interfaces to agentic action. The core idea is that AI is moving beyond being something you ask, to becoming something that does.
Marc Andrusko: The Shift from Prompting to Proactive Execution
The primary user interface for AI is moving beyond the simple prompt box. The next wave of AI applications will require significantly less manual input. Instead, they will operate by observing user context and proactively intervening with suggested actions for review and approval. This paradigm shift expands the market for software from a focus on software spend to tapping into the vastly larger market of labor spend.
The AI as an “Ultimate Employee”
The ideal AI system will function like a top-tier, high-agency employee. It won’t just identify a problem and ask what to do. It will:
- Identify a problem autonomously.
- Conduct the necessary research to diagnose its origin.
- Evaluate multiple potential solutions.
- Implement the best solution.
- Keep the user informed or request final approval.
While models are already capable of suggesting intelligent actions, the human-in-the-loop will remain crucial for final approval, especially in high-stakes environments. However, power users may eventually train their AI agents with enough context and memory to operate with near-full autonomy.
Example: The AI-Native CRM
Instead of a salesperson manually sifting through a CRM, an AI-native CRM would work continuously in the background. It would analyze emails, calendars, and past call notes to identify not just obvious opportunities but also dormant leads that could be re-engaged. The agent would then draft and suggest outreach emails, transforming the salesperson’s role from data miner to strategic decision-maker.
Stephanie Zhan: Designing for Agents, Not Humans
As AI agents become intermediaries for how we interact with systems, the principles of design must also evolve. The focus is shifting from optimizing for human consumption to optimizing for agent consumption—a concept termed “machine legibility.”
From Visual Hierarchy to Machine Legibility
For years, design has centered on predictable human behavior: good UI, intuitive flows, and visual hierarchy optimized for human eyes. As agent usage grows, this visual-centric approach becomes less critical. An agent can parse an entire document without needing a catchy headline and can process raw telemetry data without needing a Grafana dashboard. The new optimization metric is how easily and accurately a machine can read and understand information.
Content and Software in an Agentic World
- Content Creation: The old rules of journalism (e.g., hooks, the five Ws) designed to capture limited human attention are less relevant for agents. An agent won’t miss an insight buried deep in an article. This also introduces a risk: the proliferation of high-volume, low-quality content stuffed with keywords to capture agent attention, as the cost of content creation approaches zero.
- Software Interaction: Instead of engineers manually interpreting dashboards during an incident, an AI SRE can analyze telemetry data and deliver hypotheses directly. Instead of sales teams clicking through Salesforce, an agent can gather and summarize insights for them.
The key question is what agents are looking for. While the answer is still evolving, it’s clear they are better at comprehensive analysis than humans. Tools are already emerging to help organizations optimize their web presence for agent-based queries (e.g., in ChatGPT).
Olivia Moore: The Rise of Practical AI Voice Agents
AI voice agents have moved from science fiction to practical, enterprise-grade tools being deployed at scale. They are becoming the foundation for the “true AI employee” vision by handling full tasks across multiple platforms.
Enterprise Adoption Across Verticals
- Healthcare: Voice agents are used throughout the healthcare stack, from scheduling and reminders to sensitive patient-facing calls like post-surgery follow-ups and psychiatry intakes. This adoption is largely driven by high staff turnover and difficulty in staffing.
- Finance and Banking: Despite heavy regulation, voice AI excels here. It outperforms humans in compliance because it can be programmed to follow rules perfectly and its performance can be meticulously tracked.
- Recruiting: Voice AI enables instant, on-demand interviews for candidates at any time, streamlining the initial stages of the hiring process for a wide range of roles.
Technical Progress and Market Impact
Underlying models have improved dramatically in accuracy and latency; some voice agents are even being slowed down intentionally to sound more human. AI has also proven remarkably adept at handling multilingual conversations and heavy accents. The impact on call centers will be a transition where the most competitive BPOs will be those that leverage AI to offer cheaper prices and higher volume. However, in some geographies, human labor remains cheaper, delaying this transition.
Future Opportunities
The next frontiers for voice AI include government services (e.g., handling DMV calls) and consumer applications. In the consumer space, a promising area is health and wellness, with voice companions already being deployed in assisted living facilities to provide companionship and track wellness metrics.