AI Is Paying Off in the Tech Function — But Only for the 5%

See also (wiki): wiki/agentic-ai-governance.md, wiki/productivity-rcts.md, wiki/firm-size-ai-outcomes.md

Executive Summary

• BCG’s Q2 2025 global survey (n=1,250) finds AI value generated by the tech function nearly doubled year-over-year, from 7% to 13% of total enterprise AI value — second only to R&D at 15%.
• The share of tech functions scaling or fully deploying AI tripled from 9% to 28% in one year. Two-thirds now use AI in software development; 36% have reached scale.
• The value concentration is severe: only 5% of surveyed companies generate measurable AI value. Sixty percent see no material value at all. The remaining 35% are scaling but acknowledge they are not moving far or fast enough.
• Agentic AI accounts for 17% of company-wide AI value in 2025, projected to reach 29% by 2028. A third of leading companies already use agents; almost no laggards do.
• Leaders concentrate 60–70% of AI budgets on “deep agents” handling end-to-end workflows, not thin automation layers. BCG prescribes a four-tier graduated autonomy governance model for agent deployment.

Where the Value Is Concentrating

BCG identifies seven use cases where tech functions report the strongest AI adoption and measurable productivity gains. The pattern across all seven: automation of structured, repeatable work where quality and speed improvements are directly measurable.

The SDLC numbers deserve scrutiny. BCG reports 25% current productivity gains rising to an expected 44% at full scale — but does not define what “productivity” measures or whether this accounts for downstream effects on code review bottlenecks. The METR RCT (n=16, July 2025) and Uplevel study (n=800, 2024) both found that individual coding speed gains do not translate linearly to delivery throughput. BCG’s self-reported survey data cannot distinguish between perceived and measured productivity.

The legacy modernization case study is the most concrete: a financial institution used multiagent GenAI to map 3.1 million lines of code, extracting 5,000+ business rules in under three weeks — work that would have taken 7,500 hours manually. That is a genuine, measurable acceleration on a task that has historically been a black hole for IT budgets.

The Agentic Shift

BCG’s data on agentic AI adoption reveals a widening capability gap:

• Agents account for 17% of company-wide AI value in 2025, projected to reach 29% by 2028.
• The top 5% of companies allocate 15% of their AI budgets to agents. A third of these leaders already use agents in production; virtually no laggards do.
• Within tech functions, agentic AI has moved to production in nearly a quarter of companies, concentrated in service desk automation, SDLC orchestration, compliance enforcement, and infrastructure optimization.

BCG prescribes a graduated autonomy framework for agent governance — four tiers from shadow mode (agents observe, humans act) through full autonomy (agents execute, humans handle exceptions). The concept of “agent design cards” — standardized definitions of purpose, boundaries, and failure modes created before code is written — is a practical governance artifact that mid-market CIOs can implement immediately.

The platform architecture recommendation (model gateway + shared context/memory + orchestration tooling) aligns with Anthropic’s Model Context Protocol framework and OpenAI’s “Frontier” platform positioning. BCG is vendor-neutral in this section, which adds credibility.

The 5% / 60% / 35% Distribution

The most important finding may be the distribution of value realization across the full sample:

• 5% generate measurable AI value (defined as revenue/cash flow increases and process improvements from AI at scale)
• 35% are scaling but acknowledge insufficient progress
• 60% see no material value at all

This is consistent with Gartner’s finding that 72% of CIOs report breaking even or losing money on AI (n=506, 2025) and Accenture’s 8% scaling enterprise-wide (n=2,000, 2025). The tech function may be the bright spot within the enterprise, but even there, the denominator matters: when only 5% of companies generate real value, a doubling of the tech function’s share (7% to 13%) still means most organizations are not capturing it.

Key Data Points

What This Means for Your Organization

The 9%→28% tripling in tech-function AI scaling tells a clear story: the CIO’s office is where AI deployment discipline lives. If your tech function is still running pilots without a scaling plan, the competitive window is narrowing — not because AI will replace your team, but because the companies in BCG’s top 5% are compounding efficiency gains that widen every quarter.

Three decisions this data supports:

1. Pick three use cases, not ten. BCG’s leaders concentrate budgets on deep, end-to-end agent workflows. The companies spreading thin across every possible automation are the ones in the 60% seeing no value. Start with SDLC, data management, or service desk — whichever has the most structured, measurable workflows in your shop.

2. Build the platform before the agents. A model gateway, shared context layer, and orchestration standard are not overhead — they are what separates scalable deployment from “agent islands” that cannot communicate. If your agents are one-off builds tied to specific vendors, you are accumulating technical debt, not AI capability.

3. Implement graduated autonomy now. BCG’s four-tier model (shadow → supervised → guided → full autonomy) with agent design cards is immediately actionable. It gives your board a governance framework they can understand and your engineering team a promotion path they can execute against.

If this raised questions about where your tech function sits on the scaling curve, I’d welcome the conversation — brandon@brandonsneider.com

Sources

1. BCG, “How AI Is Paying Off in the Tech Function,” January 2026 (Q2 2025 survey, n=1,250 companies worldwide). URL: https://www.bcg.com/publications/2026/how-ai-is-paying-off-in-the-tech-function. Credibility: MEDIUM-HIGH — Large sample, global scope, consistent with prior BCG AI maturity work. Self-reported survey data with productivity claims based on respondent estimates, not independently measured outcomes. BCG is a consulting firm with revenue tied to AI transformation engagements; interpret directionally.

2. BCG, “Are You Generating Value from AI? The Widening Gap,” 2025. Referenced for baseline 2024 data. URL: https://www.bcg.com/publications/2025/are-you-generating-value-from-ai-the-widening-gap

3. BCG, “CIOs’ Role in AI Transformation and Productivity,” early 2025. Referenced for CIO opportunity framing. URL: https://www.bcg.com/publications/2025/cios-role-in-ai-transformation-and-productivity

Temporal tier: TIER 1 — Published January 2026 based on Q2 2025 survey data. Current and directly applicable.

Cross-reference notes: These are vendor-adjacent survey findings from a consulting firm with significant AI transformation revenue. The productivity claims (25–44% in SDLC) are self-reported by survey respondents, not independently measured. Cross-reference against: METR RCT (experienced developers 19% slower, July 2025), Uplevel study (n=800, no net throughput gain, 2024), and Faros data (98% more PRs, zero delivery improvement). The 5% value-realization rate is consistent with Gartner (72% CIOs breaking even or losing, n=506) and Accenture (8% scaling enterprise-wide, n=2,000).

Brandon Sneider | brandon@brandonsneider.com April 2026