a16zHow AI Agents Will Transform in 2026 (a16z Big Ideas)
CHAPTERS
- 0:00 – 0:31
2026 big ideas overview: UI evolution, agent-first design, and voice agents
The host frames three major shifts expected by 2026: AI interfaces moving beyond prompts, products being built for agents as primary users, and voice agents becoming mainstream. The episode is structured around three contributors, each presenting a distinct thesis tied to how work and software will change.
- •Three themes: post-prompt interfaces, agent-first product design, and voice agent adoption
- •Focus on insights from investors working closely with builders
- •Sets expectation that these are near-term, actionable shifts—not distant speculation
- 0:31 – 1:01
The death of the prompt box: AI becomes proactive in the workflow
Marc Andrusko argues the prompt box won’t remain the primary interface for AI apps. Instead, AI will observe context, propose actions, and require less manual instruction as it intervenes proactively during work.
- •Prompting decreases as apps gain context awareness
- •Agents act first, then ask for review/approval
- •UI shifts from “ask” to “review and accept” interactions
- 1:01 – 1:31
From software spend to labor spend: why the market gets ~30× bigger
The opportunity expands from hundreds of billions in software budgets to trillions in labor spend. Marc frames AI agents as systems that perform work, making the effective total addressable market dramatically larger than classic SaaS.
- •Software spend ($300–$400B) vs. labor spend (~$13T in the US) framing
- •AI agents target tasks humans do, not just software licenses
- •Recasting TAM around labor changes what “winning” products look like
- 1:31 – 2:01
AI as the ultimate employee: high-agency behavior as the benchmark
Marc uses a ‘employee agency pyramid’ to define what great AI agents should do: identify problems, research, propose solutions, implement, and only involve humans at the end. The ideal agent resembles a high-agency teammate rather than a reactive assistant.
- •Low-agency: surfaces problems and asks what to do
- •High-agency: diagnoses, explores options, implements, and keeps you informed
- •“Approval at the last minute” as the ideal human touchpoint
- 2:01 – 2:32
Human-in-the-loop vs. power users: trust, memory, and near-autonomy
As models improve and costs drop, most users will want final approval—especially in high-stakes domains. Power users may invest in training and context-sharing so agents can execute almost everything automatically, measuring success by how few approvals are needed.
- •LLMs improving on speed/cost enables more proactive apps
- •High-stakes contexts keep human approval in the loop longer
- •Larger context windows + baked-in memory increase trust and autonomy
- •Power users optimize for “tasks completed without approval”
- 2:32 – 4:10
Proactive CRM as the flagship example: continuous pipeline management
Marc describes an AI-native CRM that constantly monitors pipeline, calendar, emails, and notes to recommend or draft next actions. Instead of humans navigating dashboards, the agent surfaces the best actions and prepares them for quick acceptance.
- •Agent identifies highest-impact actions without manual CRM exploration
- •Uses historical emails/call notes to resurrect warm leads
- •Drafts outreach, schedules follow-ups, and maintains momentum automatically
- •Default mode: agent proposes; human approves the last mile
- 4:10 – 5:10
Designing for agents, not humans: the product paradigm shift
Stephanie Zhang argues that as agents mediate our interaction with apps and the web, creators must optimize for agent consumption. Visual UI and human attention tactics matter less; information structure and machine interpretability matter more.
- •Agents become the intermediary user of content and software
- •Human attention patterns (hooks, skimming) aren’t constraints for agents
- •Shift from UI/UX polish toward agent-readable structure and meaning
- 5:10 – 6:11
Machine legibility replaces visual hierarchy: dashboards to Slack summaries
Stephanie highlights how agent workflows change software interaction: agents digest telemetry or CRM data and deliver synthesized insights directly to humans. This reduces the need for humans to navigate complex interfaces and prioritizes clean, parseable data.
- •AI SREs analyze telemetry and deliver hypotheses/insights in Slack
- •Sales teams receive agent summaries instead of clicking through CRMs
- •Design goal becomes “machine legible” data and outputs
- •Interfaces move from exploration to summarized decision support
- 6:11 – 7:42
What do agents want? GEO tools, ranking games, and when humans exit the loop
The ecosystem is already forming around “showing up” in agent answers (e.g., when users ask ChatGPT for recommendations), but the optimization target is still unclear. Stephanie discusses emerging tooling and how autonomy depends on liability and accuracy thresholds.
- •Rise of tools to influence presence in AI-driven recommendations (GEO)
- •Open question: which signals agents prioritize for trust and relevance
- •Some domains already run autonomously (e.g., customer support)
- •Higher-liability workflows keep humans in the loop longer
- 7:42 – 8:48
Content explosion risk: hyper-personalization, keyword-like strategies, and noise
As content creation costs approach zero, creators may flood the web with large volumes of low-quality, highly targeted content to capture agent attention. The incentive structure shifts from crafting one great piece for humans to generating many variants for machine retrieval.
- •Agents can read everything; humans can’t—changes how content is structured
- •Hyper-personalized, high-volume content becomes feasible
- •Risk: low-quality “SEO-like” flooding aimed at agent retrieval
- •Optimization focuses on relevance signals rather than human hooks
- 8:48 – 9:19
Voice agents take up space: from sci-fi to enterprise deployment
Olivia Moore predicts voice agents will become a major interface and execution layer, moving toward the ‘AI employee’ that completes full tasks. She notes 2025 as the breakout year for real enterprise adoption and expects broader cross-platform capability next.
- •Voice agents shifting from novelty to scaled enterprise deployments
- •Platforms expand across modalities to complete end-to-end tasks
- •Voice is positioned as a key route to the “true AI employee” vision
- 9:19 – 10:50
Where voice is winning now: healthcare, finance compliance, and recruiting
Olivia details verticals already adopting voice AI at scale, led by healthcare across both back-office and patient-facing calls. Financial services adoption is driven by consistent compliance, while recruiting benefits from instant, always-available screening interviews.
- •Healthcare: insurers/pharmacies/suppliers + patient scheduling and sensitive follow-ups
- •Driver: staffing shortages and turnover make reliable automation attractive
- •Finance: AI can adhere to compliance more consistently than humans
- •Recruiting: candidates interview anytime; humans handle later stages
- 10:50 – 11:50
Operational realities: latency, “human-ness,” BPO disruption, and multilingual strength
As accuracy and latency improve, some teams even slow agents down or add noise to feel more human. Olivia discusses the uneven impact on call centers/BPOs and highlights voice AI’s strong performance with accents and multilingual speech.
- •Model improvements drive better accuracy and responsiveness
- •Some products add delay/background noise to match human expectations
- •BPO/call center shift: customers may buy “solution” providers using AI
- •Economics vary by geography where human labor may still be cheaper
- •ASR excels with accents and multilingual conversations
- 11:50 – 13:33
Next frontiers and how to build: government services, consumer wellness, and platform layers
Olivia points to government phone workflows as a major opportunity and expects more consumer voice AI, especially in health and wellness. She closes by framing voice as an industry with winners at multiple stack layers and recommends experimenting with major platforms.
- •Government services (e.g., DMV-style calls) as high-friction opportunities
- •Consumer voice: companionship and wellness tracking in assisted living/nursing homes
- •Voice AI as an industry with multiple layers of value capture
- •Builder advice: test leading platforms/models (e.g., ElevenLabs) to understand capabilities
