Agentic trust exchange: Specifications for defining agentic experiences
- commitment
- 15 hrs/week
- flexibility
- No mentee proposal
The project
We are approaching a future in which users delegate part—or even all—of their interaction experiences to digital persona agents. As end users run personalised agents on personal devices or at the CDN edge, agent-to-agent interactions governed by pre-agreed human contracts will become the norm. Over time, these contracts will serve a similar purpose to human contracts: enabling better interaction experiences. In this world, “agentic experience” will become as important as user experience, and systems will be designed with it in mind.
As cooperative multi-agent interaction becomes the next step in this sentient future, we need new paradigms for managing agentic experiences. Veripol, the platform I am building, aims to be the trust exchange for these multi-agent setups, among other capabilities, and to enable such agentic experiences.
While many elements are essential to defining such an experience, this project will clarify what good agentic experience looks like through its definitions and constructs. The aim is not to implement these constructs, but to identify them in a domain-agnostic way, grounded in a common grammar.
As part of this project, I would like to achieve the following goals:
- Develop a clear notion of what agentic experience means.
- Provide sample descriptions of what good and bad agentic experiences look like from an agent’s perspective.
- Complete three use cases across three different domains to capture both domain-agnostic and domain-specific constructs in each envisaged agentic contract.
- Draft sample contracts that can be written across a fleet of agents participating in a given use case.
The mentee's role
Mentees will own the research process and think from first principles, given the transformative impact of this technology. Areas they will work on may include:
- Study examples of great user experience stories and identify equivalent underlying concepts that help define agentic experiences.
- Research and potentially write a paper on what a good contract between a human and an agent looks like.
- Extrapolate how a human-defined contract could influence other agents working for their corresponding human owners across three hypothetical scenarios.
- Identify which parts of a contract are domain-agnostic versus domain-specific.
- Implement sample agentic applications to simulate agent interactions and identify what could contribute to good and bad agentic experiences. The goal of the code is not to implement a solution, but to create a test bed that clarifies why agentic experience matters for the use cases identified.
- Define the foundations of a good agentic experience, as well as what could lead to bad agentic experiences.
Ideally, mentees will contribute to a paper for arXiv, though this is not a primary goal.
Who I'm looking for
Must-haves:
- Ability to conduct independent research and share a few ideas on a topic within a week (or within an agreed time period).
- Ability to implement sample agentic applications (ideally in Python; TypeScript skills are also helpful).
Nice-to-haves:
- Experience with product design and iterative improvement.
- Experience designing public API specifications.
- Experience designing human–computer interfaces.
- Domain expertise beyond coding (e.g., finance, marketplaces, compliance, risk).
Questions for applicants
Imagine a near future in which humans have a collection of “personal agents,” and you have your own fleet of agents to enhance your everyday life and work. What would your fleet look like, and how would you want those agents to interact with other agents (whether another agent in your own fleet or an agent belonging to other humans)?
- Feel free to think out loud about what this future could look like, and use an example or two to clarify.
- Ideally fewer than 1,000 words; a high-level thought process is fine.
Support offered
- Participate actively and contribute to the design process
- Facilitate connections within the team
- Be available on Slack for offline communication, and hold 1:1s for individual guidance
- Help design sample applications
- format
- Product & entrepreneurship
- topic
- Post-AGI transition · Artificial sentience · Other
Ananth Gundabattula
I am building an AI assurance platform for the AI workforce called Veripol. Veripol aims to enable trust and assurance for the AI workforce of the future through capabilities such as activation monitoring, digests, and agentic personas.
I am a technology leader with 23+ years of experience building large-scale distributed systems, ML and data platforms, agentic frameworks, and fraud/cybersecurity infrastructure. Previously, I co-founded Darwinium (Series A, $26M raised). I hold a PhD in Computer Science (Security, 2005–2010). I am a named inventor on three patents, a member of the IBM Patent Review Board, an Apache Apex committer, and a published co-author in the Springer Encyclopedia of Big Data Technologies. I have spoken at Strata San Jose, YOW! Data, and Dataworks Summit.