Agentic AI in ESG: The 5 Use Cases Already Reshaping Reporting

Renato Souza, executive manager at OTA, Omni Helicopters International’s parent operation, used to lose a full month each year to the company’s CDP climate disclosure. Last year, OHI tried something different. It deployed an agent-based platform built by Gardenia Technologies that queried internal databases, documents and public sources simultaneously, then assembled draft answers to CDP’s 450-plus questions. Reporting time fell to a single week. Agentic AI in ESG is still early, but “early” no longer means “theoretical.” It means companies like OHI are shipping results while competitors are still scoping pilots.

The timing matters. On 3 July 2026 the European Commission adopted revised sustainability reporting standards, cutting mandatory data points by over 60%. Omnibus I narrowed CSRD scope to roughly 5,000 companies, down from 50,000. But even simplified rules cost serious money. EFRAG estimates recurring compliance at a minimum of €1.37 million a year for large firms, often two to four times that. The next wave reports in 2028 on FY 2027 data. Eighteen months is not a lot of runway to build something from scratch.

1. Agentic AI in ESG disclosure drafting

OHI’s deployment is the clearest proof point. Gardenia’s Report GenAI, running Claude on Amazon Bedrock, uses three specialised agents working in parallel: one handles structured database queries, another retrieves from internal documents, a third pulls regulatory context from public sources. A synthesis agent stitches everything into framework-compliant answers. Of the 97% of questions the system pre-filled, OHI’s sustainability team approved 70% without editing a word. Each answer carried source citations and a reasoning trace, so reviewers could interrogate the logic rather than trusting the output blindly.

Now scale the problem up. CDP runs to over 450 questions covering climate risk, water stewardship and energy consumption. A full CSRD filing spans governance narratives, transition plans, risk assessments and quantitative metrics across multiple ESRS topical standards. Nobody has demonstrated an agent system doing all of that end-to-end yet. But BCG’s 2026 white paper on AI-enabled ESG reporting identified disclosure drafting as the use case with the most measurable efficiency gains, and OHI’s experience shows why: it is structured, repetitive and data-intensive, which is precisely where agents outperform humans.

2. Scope 3 carbon estimation

If disclosure drafting is the most immediately practical use case, Scope 3 is the most consequential. These indirect emissions, buried in supply chains, business travel and downstream product use, make up the majority of most companies’ carbon footprints. They have also been effectively unmeasurable at scale because the data lives across hundreds of suppliers, procurement platforms and expense systems that were never designed to talk to each other.

The revised ESRS changed the equation by permitting estimates, proxy data and sector-average benchmarks instead of demanding direct supplier figures. That regulatory concession is what makes agentic AI in ESG carbon accounting more than a nice idea. Vendors like Carboledger and Normative now connect directly to ERP systems, match transactions against current emission factors and update the numbers continuously as spend data flows through. No quarterly spreadsheet exercise. No months of chasing supplier emails.

Auditors do not need Scope 3 numbers to be precise. They need them to be defensible: methodologically consistent, timestamped, traceable. That is what the ESRS explicitly permits under its “without undue cost or effort” principle, and it plays directly to what agents are good at.

3. Supply chain ESG screening

CSRD-obligated companies must disclose across their value chains, which means their suppliers’ sustainability credentials are no longer a nice-to-know. A lapsed environmental certification or an unreported labour violation at a tier-two supplier becomes the reporting company’s problem.

Doing this manually means monitoring thousands of vendors across certifications, labour standards, human rights records and emissions profiles. It does not scale with headcount. General Mills took a different approach, deploying an AI-driven supply chain system that evaluates over 5,000 daily shipments. Exceptions get escalated. Routine decisions happen autonomously. The system has generated more than $20 million in savings since fiscal 2024, though that figure covers logistics optimisation broadly, not ESG screening specifically.

What makes agentic AI in ESG supplier work particularly interesting is a constraint that Omnibus I introduced: a value chain cap limiting what large companies can demand from smaller partners. Requests to suppliers with fewer than 1,000 employees must stay within the voluntary VSME standard. Agents that can squeeze maximum insight out of limited data, using proxy estimates and sector averages where direct disclosure is unavailable, become the only realistic way to meet both sides of that regulatory bargain.

4. Agentic AI in ESG greenwashing detection

On 27 September 2026, enforcement begins on the EU’s Empowering Consumers for Green Transition directive. “Eco-friendly” and “carbon neutral” become restricted terms without certified proof. Fines reach 4% of annual turnover. That deadline has focused attention.

Multi-agent systems now split detection into stages: one agent extracts claims from reports and marketing copy, another checks them against reported performance, a third assesses consistency across documents. Some use NLP models trained specifically on environmental language to catch phrasing that sounds green but commits to nothing. The market for these tools is projected to reach $3.14 billion by 2030, though that figure should be treated as a projection rather than settled market data.

The real adoption driver is not enforcement against competitors. It is self-protection. Companies are running their own disclosures through detection tools before publication because it is cheaper to rewrite a claim in draft than to defend it after a penalty notice. To be sure, the ICAEW raised a fair concern in April 2026: a DRCF report identified seven compliance risks when businesses deploy AI agents, including fragmented accountability. A detection tool whose reasoning cannot be audited creates a different kind of greenwashing problem, not a solution to the existing one.

5. Compliance orchestration across frameworks

Most companies do not file against a single standard. CSRD sits alongside CDP, TCFD, GRI, ISSB and a growing list of jurisdictional requirements. The evidence is often the same. The mapping is not. Without orchestration, compliance teams end up doing the same work three or four times for different frameworks, which is where costs spiral.

Vanta, now at $300 million in annual recurring revenue and a Forrester-recognised Leader in GRC, maps controls across more than 35 frameworks including CSRD, DORA and the EU AI Act. Goldman Sachs approached the problem differently, embedding Anthropic engineers for six months to build agents for trade accounting and regulatory compliance. CIO Marco Argenti told CNBC the model handled areas that had resisted automation for decades. Internal tests showed 30% faster onboarding and 20% developer productivity gains, with over 12,000 developers now using the system. Self-reported numbers, but the underlying logic applies broadly: compliance orchestration is high-volume, rules-based and repetitive. It is the use case where agentic AI in ESG delivers the widest margin over manual work.

Where agentic AI in ESG goes from here

Gartner predicts 40% of enterprise apps will feature task-specific agents by year-end, up from under 5% in 2025. In ESG, the adoption curve is being compressed by a hard deadline. Companies filing CSRD disclosures on FY 2027 data have roughly 18 months to build the data collection, materiality assessment and assurance infrastructure the standards require.

None of the deployments above prove that agent systems can handle a full CSRD filing today. What they show is that the individual components of agentic AI in ESG, from disclosure drafting and carbon estimation to supplier screening and framework orchestration, work in production for defined scopes at named companies. BCG’s white paper made the point directly: the bottleneck is not tool availability but structural integration into reporting workflows. That integration work is happening now, quietly, at the companies that expect to file on time. Everyone else is watching.