Big Ideas 2026: The Enterprise Orchestration Layer

Seema Amble explains how enterprises must shift from isolated AI tools to coordinated multi-agent systems that behave like digital teams. The Fortune 500 will need to extract tacit knowledge from people's brains, documents, and actions to create shared context that enables agents to collaborate across departments with proper feedback loops and KPIs.

• •Context extraction combines documentation (onboarding videos, written instructions) with watching human actions (browser clicks, phone calls)
• •Agents need cross-functional feedback loops - e.g., support agents telling sales agents which customers are unprofitable
• •Multi-agent failures are prevented through eval functions, KPIs per agent, and humans-in-the-loop at key decision points
• •Fortune 500 companies have the most siloed data across acquisitions and geographies, making context layers especially valuable
• •New context layers can reduce ERP/procurement implementation from years to much faster timelines

Angela Strange argues 2026 is the turning point where financial institutions will replace legacy systems because the risk of not changing exceeds the risk of change. Next-gen AI-native platforms unify data from legacy cores and external systems, enabling parallelized workflows and dramatically expanding software categories.

• •Three drivers: mainframes breaking under scale, revenue left on table (underwriters can't process demand fast enough), and viable AI-first alternatives now exist
• •Workflows become parallelized - mortgage teams can execute 400+ tasks in parallel with agents handling mundane work
• •Categories expand - customer data from onboarding, KYC, transaction monitoring, and service interactions unified into single risk platforms
• •Winners will be 10x bigger by consuming labor humans didn't want to do and couldn't hire for fast enough
• •Early adopters turning 5% margin businesses into 50% margin businesses in areas like mortgage servicing

Alex Immerman describes vertical AI evolving through three phases: information retrieval, reasoning (2025), and multiplayer mode (2026). Complex work requires collaboration between multiple humans and agents with explicit trust rules, creating command center interfaces that separate autonomous execution from human review.

• •Vertical AI companies growing faster than historical SaaS because they can replace specialized labor better than horizontal tools
• •Multiplayer mode increases platform value and switching costs, addressing defensibility concerns about vertical apps
• •AI operating agreements define when agents can act autonomously vs. when they must flag issues to humans
• •Example: M&A agents negotiate within set parameters (min/max price), flag contingencies and earn-outs for human decision
• •Software becomes command centers with two sections: activities agents execute autonomously and flags requiring human engagement
• •Defensibility comes from brand (high referenceability in vertical markets), proprietary technology, and network effects from multiplayer adoption

David Haber argues the strongest AI companies are those where AI reinforces the business model by driving revenue and outcomes, not just reducing costs. Market pull is much stronger when AI enables customers to make more money rather than just save on expenses, with defensibility coming from end-to-end workflow ownership and proprietary outcomes data.

• •Eve (plaintiff law): AI enables attorneys on contingency to take on more clients and make more money, not eroding billable hours
• •Salient (loan servicing): Voice agents in 50 languages not only reduce call center costs but drive better collection rates
• •Defensibility through end-to-end workflow ownership from intake to outcome, embedding deeply in daily customer workflows
• •Proprietary outcomes data creates compounding advantage - not available on public internet for model training
• •Eve uses outcomes data to inform smarter intake decisions ($50K vs $5M case potential) and optimize demand letters based on counterparty characteristics