Tessl today outlined a plan to extend a registry it created for specifications that govern the behavior of artificial intelligence (AI) agents being used to build software to create a platform for creating, distributing and observing those specifications.
Speaking at a DevCon Fall 2025 conference hosted by the company, Tessl CEO Guy Podjarny said that as a specification approach to using AI agents to build software gains traction, the next major challenge is finding a way to share context and knowledge across the multiple AI agents that DevOps teams will be relying on. A Tessl Agent Enablement Platform will make it possible to achieve that goal by capturing context and knowledge in the form of a specification file that can be easily shared across AI agents, he added.
A specification-based approach to AI coding promises to both improve the quality of the code created by AI coding tools and scale beyond a single application developer, noted Podjarny. Specifications provide the rules and context needed to more narrowly focus AI agents on a specific task. That curated approach helps prevent the AI agent from making a foolish mistake or, more troubling still, telling a developer a task has been completed when in fact it has not. Tessl previously created a framework and registry for capturing those specifications, which it is now moving to expand to provide a platform that can generate and assess specifications and then distribute to AI agents, said Podjarny.
Additionally, the Tessl Agent Enablement Platform will also enable DevOps teams to observe how well AI agents are invoking various specifications that are stored in the Tessl registry, he added. There are already more than 10,000 specifications for AI agents stored in that registry, which essentially provide AI agents with access to knowledge that is stored in a set of easily accessible files that can be governed by DevOps teams, noted Podjarny.
It’s still early days so far as adoption of AI agents that are capable of building and deploying software is concerned, but it’s clear application developers are spending a significant amount of time trying to optimize prompts by providing the context an AI agent needs to ensure the outcome desired. A specification-based approach reduces that toil by making it easier to reuse context across multiple stages of an application development workflow. The more narrowly defined those stages are the less chance there is that the AI agent will generate an erroneous output. Otherwise, the AI agent, while being endowed with superpowers, is also going to be massively unreliable, said Podjarny.
There is little doubt that in time multiple approaches to managing specifications for AI agents will emerge. In the meantime, the number of AI coding tools that support specifications to improve the quality of the code being generated continues to increase. As the latest version of these tools are adopted, the need to find ways to share specifications across teams of developers will become more apparent, noted Podjarny.
In fact, there will soon be untold numbers of AI agents embedded into application development workflows. The issue that needs to be addressed now is how to govern them in a way that ensures they don’t become more trouble than they might ultimately be worth.
