AI Enterprise Adoption

State of play.

• AI vendor pricing is restructuring enterprise contracts in real time. Salesforce, Workday, and OpenAI are abandoning per-seat licensing for consumption-based models tied to work output — "agentic work units," "units of work," tokens — with measurement methodologies still largely undefined across the sector .
• Palantir's integrated data-plus-AI thesis is under direct competitive pressure. CEO Alex Karp is publicly attacking commodity AI outputs as "slop" while investors question whether enterprises will pay Palantir's premium over cheaper standalone LLMs — even as the company raises full-year guidance to $7.2B on 61% projected growth (→ Palantir CEO Karp slams AI "slop" amid fears of losing business to rival models).
• The enterprise AI architectural shift is accelerating from chatbot-first to embedded infrastructure. McKinsey, Deloitte, and Microsoft research documents that organizations redesigning core processes around persistent, governed AI — rather than bolting tools onto legacy workflows — are the ones achieving scale; Anthropic and IBM are formalizing this through context engineering and runtime governance guidance .
• Shadow AI adoption is endemic and governance frameworks are lagging. A 2025 Gartner survey found 69% of organizations suspect or have confirmed unsanctioned AI use; the figure reaches 98% when counting all applications; 93% of executives report using unauthorized AI themselves .
• For counsel advising enterprise clients, law firms, or AI vendors, the practical baseline is: consumption-based pricing is arriving before contract terms are standardized, embedded AI infrastructure creates audit-trail and accountability structures that existing governance frameworks do not yet address, and the Palantir debate crystallizes the build-vs.-buy and vendor-lock-in questions clients will be asking in the next procurement cycle.

Where things stand.

• Consumption-based AI pricing is displacing per-seat licensing. Salesforce charges for "agentic work units," Workday for "units of work," and OpenAI signals a shift toward token-based utility pricing — confirmed by a Goldman Sachs analysis of roughly 40 software and internet companies . Contract terms and measurement methodologies remain undefined, creating immediate drafting exposure for procurement counsel.
• The enterprise AI architectural model is shifting from visible tools to embedded infrastructure. Research cited by McKinsey, Deloitte, and Microsoft shows 89% of organizations deploy AI somewhere, yet only approximately 33% have achieved meaningful scale — the gap explained by organizations that bolt AI onto legacy systems rather than redesigning workflows around intelligence as infrastructure. Anthropic and IBM are formalizing embedded-system governance through context engineering and runtime governance guidance . This shift creates different audit trails, accountability structures, and failure modes than user-facing tools — and existing AI governance frameworks do not yet address agent autonomy, decision lineage, or human oversight in embedded contexts.
• Enterprise AI pilots are failing at scale despite massive investment. The culprit is organizational and cultural, not technical — data architecture, governance gaps, workflow misalignment, and change management failures dominate . Leadership alignment, not tool capability, is the determinative factor: Microsoft's Work Trends Index — 20,000 users across 10 countries — found organizational factors have twice the impact of individual factors on successful AI integration; only 25% of AI users perceive their leadership as clearly aligned on AI strategy; and only 13% of employees report being rewarded for reinventing their work .
• Shadow AI adoption is endemic and governance frameworks are lagging. A 2025 Gartner survey found 69% of organizations suspect or have confirmed unsanctioned AI use; the figure reaches 98% when counting all applications; 93% of executives report using unauthorized AI themselves . One-third of employees admit to sharing enterprise research or datasets through unsanctioned tools, 27% have exposed employee data, and 23% have input company financial information into these platforms — creating data breach, regulatory, and IP exposure across healthcare and financial services .
• Law firm AI adoption is bifurcating by size and pricing model. Specialized legal AI platforms deliver documented returns — a GC AI study of 100+ customers found 14 hours per week saved per lawyer and 14% reduction in outside counsel spend . Clio's 2026 Legal Trends report documents the small-firm problem: 71-75% AI adoption, fewer than 33% revenue growth, 86% still on hourly billing . BigLaw is institutionalizing AI at the firm level — Mayer Brown has mandated generative AI training for all 1,800 lawyers globally; Goodwin Procter has committed to a 90% daily usage target .
• Palantir's integrated data-plus-AI platform faces commodity-LLM competition. Karp's "slop" framing sharpens the enterprise vendor-selection debate; critics point to vendor lock-in through "black box" code; CTO Shyam Sankar counters that AIP drives job creation through factory efficiency gains (→ Palantir CEO Karp slams AI "slop" amid fears of losing business to rival models). Palantir has raised full-year 2026 guidance to $7.182–$7.198B, projecting 61% year-over-year growth, with US commercial revenue projected to climb over 115% .
• Sector-specific AI agents are entering industrial and procurement workflows. Emanate's AI agents compress industrial materials quoting from 3-4 weeks to near-instant, with 8-12 week implementation cycles and revenue-growth targets of 40%+ per client — a pattern of sector-specific deployment appearing across manufacturing, logistics, and supply chain .
• Capital formation around AI infrastructure and deployment remains at velocity. Google has committed up to $40B in Anthropic . Wall Street is sorting software companies into AI winners and losers, with horizontal SaaS incumbents under pressure .
• Change management is the implementation bottleneck. SimplePractice's CLO ran a hands-on team exercise to shift employee perception from fear to innovation — a bottom-up approach that contrasts with top-down mandates and reflects the broader finding that psychological safety and experimentation culture drive adoption .

Latest developments.

• Enterprise organizations are shifting from chatbot-first to embedded AI infrastructure — McKinsey, Deloitte, and Microsoft research documents the architectural transition; Anthropic and IBM are formalizing context engineering and runtime governance guidance; embedded systems create audit-trail and accountability structures that existing governance frameworks do not yet address
• Palantir CEO Karp publicly attacks commodity AI outputs as "slop" while investors question whether enterprises will pay for integrated data-plus-AI over cheaper standalone LLMs; vendor lock-in and "black box" code complaints resurface (→ Palantir CEO Karp slams AI "slop" amid fears of losing business to rival models)
• Emanate launches AI agents for industrial materials quoting — a16z-backed startup compresses sales cycles from weeks to near-instant in the metals and minerals sector, measuring success on 40%+ revenue growth targets rather than cost reduction

Active questions and open splits.

• Embedded AI governance: audit trails, decision lineage, and human oversight. The shift from user-facing chatbots to persistent, invisible infrastructure changes every assumption in existing AI governance frameworks — who is accountable when an embedded system makes an autonomous decision, what constitutes an adequate audit trail, and whether current compliance frameworks map onto systems that operate without a visible human-in-the-loop are all open .
• Consumption-based pricing: measurement and cost-cap terms. The shift from per-seat to work-output billing is moving faster than contract standards. How "agentic work units" and "units of work" are defined, audited, and capped is unresolved — and vendors are setting terms before enterprise procurement teams have frameworks to push back .
• Integrated data-plus-AI vs. commodity LLM: the Palantir question. Whether compliance-first, ontology-based platforms justify premium pricing over faster, cheaper generic LLM deployments is the question clients in regulated industries will be asking procurement and outside counsel. If the premium erodes, existing Palantir contracts face renegotiation pressure; if regulators tighten AI governance, Palantir's positioning becomes a competitive advantage (→ Palantir CEO Karp slams AI "slop" amid fears of losing business to rival models).
• Shadow AI governance: block, monitor, or channel. The data makes blocking unrealistic — 98% penetration including C-suite. Channeling requires governance infrastructure most organizations have not built, and one-third of employees are already sharing enterprise data through unsanctioned tools. Whether deliberate misuse constitutes a compliance failure or an employment-performance issue is unsettled .
• Law firm billing model under AI pressure. The performance paradox — firms capturing productivity gains while leaving pricing models unchanged — is documented across Am Law 100 and small-firm cohorts alike. Whether client demands for AI-efficiency discounts will force structural fee-arrangement changes, and whether firms that raise rates without demonstrating AI value face client attrition or malpractice exposure, is the open question for firm management .
• Leadership accountability for AI outcomes. Microsoft's research frames AI failure as a leadership problem, not a technology problem. Whether boards and executives face fiduciary or duty-of-care exposure for AI adoption failures — particularly where governance frameworks were not embedded at pilot inception — is an emerging question without settled doctrine .
• Sector-specific agent deployment: liability allocation as AI moves autonomous. Emanate's model — AI-generated quotes initially under human review, transitioning to fully autonomous operation as client trust builds — is the pattern across industrial AI deployments. The contractual and liability questions around that transition point, and who bears responsibility when autonomous outputs are wrong, are not yet standardized .

What to watch.

• Whether Anthropic's joint venture with Blackstone and Goldman Sachs discloses governance terms, liability allocation, and Claude deployment contracts — these will become reference points for the next wave of AI-lab/enterprise deals.
• Whether Palantir customer churn accelerates over the next two quarters as enterprises evaluate commodity LLM alternatives — and whether any renegotiation or migration disputes surface publicly.
• Whether consumption-based pricing disputes surface in litigation or arbitration as enterprises discover that "agentic work unit" definitions were not adequately defined at contracting.
• Whether Anthropic's or IBM's context engineering and runtime governance guidance for embedded AI systems becomes a market-standard reference point for enterprise AI governance frameworks — and whether regulators adopt or reference it.
• Whether additional major law firms follow Mayer Brown's mandatory AI training model or Goodwin's AI-native target — and whether bar associations begin issuing competency guidance that references specific adoption thresholds.
• Whether any organization publishes a shadow-AI governance framework that becomes peer-standard — the current vacuum remains the most immediate compliance gap across the cluster.