Nesta’s residency programme is designed to promote knowledge transfer, foster continuous learning and help to build new networks in the sectors in which we work. Simon Wisdom is our agentic AI resident from September to December 2025, embedded within Nesta's Discovery team but working across missions. He is spending a few months exploring a range of agentic AI applications.

Simon’s primary project falls under Nesta’s sustainable future mission, in partnership with Renbee, a startup tackling home decarbonisation for installers and consumers. Heat pump installations are critical for UK net-zero targets, but are severely delayed by administrative bottlenecks. Simon is working with them to build a distribution network operator (DNO) communications manager agent to handle challenging and time-consuming parts of the application. There is an opportunity to expand the capabilities of the agent to do even more on Renbee’s platform in the future by giving it additional tools.

Working alongside the healthy life mission, Simon’s other main project focuses on the adoption of the proposed healthy food standard, which will require retailers to report data on the nutrition profiles of their product sales. Auditing this data will be a massive task, and there is a big opportunity to incorporate agentic AI into this process to streamline the process of identifying data anomalies and non-compliance, making it easier for the government to adopt and enforce the standard.

Alongside these two headline projects, Simon is also building a few small prototypes to demonstrate value for internal tasks at Nesta.

We want to show how agentic AI can bring value to Nesta’s missions. We plan to generate several proof of concepts that are promising enough to develop into ongoing projects.

For the DNO communications manager, we would consider it a success if we successfully integrate the agentic tool on Renbee’s platform, including metrics that start to measure time savings on the application process.

For the product nutrition audit, we want to demo to senior decision-makers in the government that AI-assisted compliance auditing is solvable at scale.

Across both projects, and a few other small prototypes, we want to create replicable open-source patterns for agentic AI implementation that Nesta and other organisations can adopt and scale.

Across Nesta’s missions, a range of administrative bottlenecks and time-consuming and repetitive processes slow down progress towards our goals. Scanning these domains, we have identified a set of problems that we believe agentic AI could solve. 

For the sustainable future team, there are a number of administrative challenges that heat pump installers must navigate when they install a new heat pump.

In particular, the DNO application process causes significant delays and headaches for installers. Anecdotally, we have learned:

• many DNO applications are rejected on the first submission and often require additional information from the installer
• delays of 3-11 days are common, mostly waiting for trivial forms or confirmation emails
• DNO applications consume a non-trivial amount of installer’s time (definitely enough to cause delays and frustration).

If agentic AI can reduce this burden, then we can free up installers to spend more of their time actually installing heat pumps.

Similarly, Nesta’s healthy life team is grappling with how the government’s healthy food standard can effectively collect and analyse data for enforcement. To support this ambitious plan, we are prototyping how agentic AI could help audit reported data and enable enforcement at scale.

DNO communications manager

An email agent that manages communication between heat pump installers and DNOs. This is an agentic workflow, built with LangGraph, that monitors DNO responses to installers , proactively collects relevant information, interprets what is being asked, follows-up and submits forms on behalf of an installer. 

This is a full end-to-end project, with the agent workflow hosted and run on cloud infrastructure, communicating with Renbee via an API, accessing relevant email communications, and collecting human-in-the-loop feedback from installers. We can’t share all the details of the implementation, but we plan on writing up a lot of our learnings about the process and the framework.

Nutrition database auditor

An audit agent that investigates tabular product data using a toolkit (eg, calculation, cross-referencing, web search) that mimics a human auditor. We are creating a test dataset by injecting synthetic "nefarious" and “benign” anomalies into an existing dataset for the agent to investigate. The agent's logic will encode expert workflows to provide a reasoned justification for classifying anomalies as "nefarious gaming" or "innocuous errors." The output will be presented on a dashboard that non-experts can understand and query.