Gravity launches AI sustainability agent & buyer guide

Gravity has launched an AI agent for sustainability and energy teams, designed to handle any action a user can take within its platform.

The San Francisco-based carbon and energy management software provider says the new Gravity Agent includes more than 85 prebuilt skills developed by its in-house climate strategists and energy specialists. These cover tasks such as migrating historical data from spreadsheets and other platforms, drafting regulatory disclosures, and identifying energy cost-saving projects.

Gravity is also publishing a buyer's guide to help software purchasers assess vendors' claims about AI agents. The guide sets out nine questions for buyers, including whether a product can do everything a user can do in the software or only the tasks shown in a demonstration.

Ted Kornish, co-founder and Chief Technology Officer at Gravity, argued that the term has become too loosely applied across the software market.

"Everyone says they have an agent, but almost no one means the same thing," Kornish said.

He said Gravity is trying to distinguish between software that can carry out a narrow task and software that can complete broader workflows.

"A true agent handles a wide variety of tasks and works toward them: deciding the steps, taking action, and circling back only when it needs more guidance or has a draft ready for review. Gravity is the only API-first sustainability platform, meaning the agent can use the software in all the ways users can," he said.

Buyer scrutiny

The buyer's guide reflects a wider debate in enterprise software over what should qualify as an AI agent. Gravity's framework distinguishes between systems that answer questions about data, tidy files, or draft reports, and systems that can take a goal, plan steps, and carry out actions across a platform before returning for approval.

It recommends that buyers ask vendors to use a customer's own data during a demonstration, test the software with unscripted tasks, and run a short pilot on a real business problem. It also urges buyers to ask how the software handles permissions, whether every action is logged, and whether calculations are generated through fixed formulas rather than by a language model.

For Gravity, the product launch and the guide are linked. The company is positioning its software against what it describes as "agent-washing", a term it uses for products branded as agents despite being limited to a small set of prepared tasks.

How it works

Gravity says the Gravity Agent can work on multiple tasks at the same time, operate in the background, and retain memory from prior work. Users can also create new skills tailored to their needs, with support from Gravity's climate strategists.

The software includes safeguards because carbon accounting and energy management are subject to scrutiny from internal teams, regulators, and auditors. Gravity says the agent cannot make changes without human review and approval, and that every action is recorded to preserve an audit trail.

Customer data is not used to train the agent, according to the company. Emissions calculations continue to run through Gravity's deterministic calculation engine, which it says makes outputs reproducible for customers and their auditors.

Early results

In early pilots, Gravity says customers reduced the time needed to migrate data, prepare disclosure drafts, and identify energy cost savings by 70% or more. The company did not name the pilot customers, but said the reductions turned work that had taken days or weeks into tasks completed in hours or minutes.

The launch comes as sustainability teams face growing workloads tied to emissions reporting, energy management, and compliance requirements. Many organisations still rely on spreadsheet-based processes and manual review, making the promise of workflow automation attractive if it can be trusted in regulated settings.

Gravity says its skills library has been trained and validated by subject specialists rather than relying only on a general-purpose foundation model. The company argues that this expert input is necessary in sustainability and energy work, where reporting standards, emissions factors, and auditor expectations can change frequently.

The business also pointed to its earlier work on measuring the emissions associated with AI usage. It said that methodology is now used to account for emissions from the Gravity Agent itself and to help customers measure emissions linked to their own use of AI.

Gravity's central claim is that an AI agent should be judged less by its label than by whether it can perform real work inside software under the same controls as a human user. "Before you sign, make a vendor prove it on your data, not on slides," Kornish said.