Talk to us

R&D Lab

The research behind agents that hold up in production.

The lab researches how to make enterprise AI agents work reliably, improve over time and operate at a sustainable cost. Our questions come from live deployments, and the answers that hold up go back into how we build.

Dr. Ayesha Khanna, co-founder of Addo AI, smiling as she leans on a railing by the water
Our R&D lab, led by our co-founder Dr. Ayesha Khanna, develops and tests the next generation of enterprise AI capabilities.

Six research areas.

Each area starts with a failure we have seen in production and ends with a method we can test on client workflows before it ships.

Reliable computer-use agents

Many enterprise systems have no API, so agents must work through the software itself. We research how agents verify their actions, recover from interrupted workflows and prevent duplicate submissions, even when applications give incomplete or ambiguous feedback.

Action verificationFailure recoveryLegacy workflows

Enterprise evaluation

A convincing demo is not proof of dependable performance. We research which evaluation methods best predict real-world reliability, using representative client workflows to measure task completion, policy compliance and behaviour under unexpected conditions.

Workflow evalsShadow testingRelease gates

Learning from operations

Every reviewer correction reveals something about how an agent fails. We research how to turn operational feedback into measurable improvements in knowledge, instructions and tool use, without regressions in workflows that already perform reliably.

Reviewer feedbackTargeted improvementsRegression testing

Agent planning and task decomposition

Complex enterprise tasks require agents to plan, adapt and coordinate many steps. We research when agents should reason dynamically, follow structured workflows, or combine both to complete tasks reliably and efficiently.

Task decompositionAdaptive planningExecution strategies

Agent memory and context engineering

Enterprise agents need the right information at the right time. We research how memory and context strategies affect accuracy, continuity and reliability across long-running workflows, especially when information is outdated, incomplete or contradictory.

Context engineeringAgent memoryState management

Agent efficiency and cost optimisation

More reasoning and larger models do not always produce better outcomes. We research how model selection, reasoning strategies and tool-use patterns affect cost, latency and task success, and where simpler approaches deliver comparable reliability.

Model routingTool efficiencyCost and accuracy trade-offs
Two researchers sitting with a bank clerk at her desk, watching how she works

Research in the field.

Our researchers sit with the people who do the work, inside the client's operations, before they write a line of code.

Researchers sit with deployment teams.

The lab is part of every engagement.

Live deployments raise research questions; the lab turns them into methods tested on client workflows; the methods that hold up become how we build every agent, which goes back into live deployments. Applied researchers work in every deployment. Live deployments Research questions Methods tested onclient workflows How we buildevery agent Applied researchersin every deployment
ProblemsResearch questions come from live deployments, ranked by how many clients they affect.
PeopleApplied researchers rotate into client teams as part of every deployment.
PlatformResults that hold up across clients become part of how we build every agent.
LanguagesFrom offices in San Francisco, Singapore, Dubai and Lahore, we work on agents that handle Arabic, Urdu, Bahasa, Malay and code-switched conversation as well as they handle English.
PrivacyClient data stays with the client. Shared improvements use methods, not data.

Writing on AI, work and policy.

Addo AI's leaders write and speak on AI, data and the future of work, including for the World Economic Forum.

World Economic Forum

Preparing people for the AI workplace

What education systems and employers need to change as AI takes on more of everyday work.

World Economic Forum

Regulating the sharing economy

How governments can regulate data-driven platforms while keeping room for new services.

Have a problem no template solves?

Bring us the workflow where agents keep failing. Our researchers and engineers will tell you why, and what it would take.

Talk to our researchersJoin the R&D Lab