Agent Harness Explained with Tero

Platforms for creating custom AI agents are becoming increasingly common, along with shared repositories where teams can access prebuilt agents and use them consistently.

They solve a real problem: centralizing agent creation and access. But for software quality, that isn’t enough. What business context does the agent reason with? Which systems can it access? What evidence does it leave behind? And who reviews its decisions before they reach production?

Tero is Abstracta’s answer to these questions. It doesn’t compete as a general-purpose agent platform. It’s an agent harness designed specifically for QA workflows.

Tero is also the technological core of Tero Enterprise Edition, one of the three components of Abstracta Intelligence, Abstracta’s solution for bringing AI into software quality across the development lifecycle.

Why This Matters Now¶

According to a Gartner report, IT spending in banking and investment services is projected to exceed USD 1.1 trillion by 2029. But increasing investment and adding AI tools alone does not solve how teams make quality decisions.

AI can make individuals more productive. But at the team level, people are still critical for understanding complex systems, navigating compliance requirements, and applying the expertise of testers and analysts.

Turning individual productivity gains into team capability requires shared context, reusable knowledge, and evidence to support production decisions.

That’s why Tero isn’t designed as an individual productivity tool, but as an agent harness: a layer that turns this knowledge into a governed, measurable capability shared across the team.

What Makes Tero Different from Other Platforms that Also Let Teams Build and Share AI Agents?¶

Platforms for building and sharing agents vary in focus: some work with production data, while others are SWE Agents (Software Engineering Agents) designed to handle programming tasks. Tero, by contrast, focuses specifically on agents for software quality and isn’t designed to compete as a general-purpose agent platform.

Its key differentiator is being an agent harness specialized in QA workflows and in the context organizations need to operate those agents consistently, with governance and integration into real software delivery processes. This includes testing and automation, as well as functional analysis, system understanding, knowledge management and reuse, collaboration across QA, development, product, and business teams, security, governance, and integration with the organization’s technology stack.

What Does It Mean That Tero Is an Agent Harness?¶

An agent harness is the software layer around an AI model that defines how it operates for a specific task: what context it receives, which tools it can access, what rules it follows, and what output it produces.

In Tero, the model provides the reasoning; Tero provides the workflow, context, controls, integrations, observability, and evidence needed to apply that reasoning consistently to QA.

That’s why Tero goes beyond hosting or sharing agents. It defines the system in which those agents operate: their objective, the information they use, the systems they can access, the actions they’re allowed to perform, the evidence they must collect, the process they must follow, when human review is required, and the artifact they must produce.

This is the main difference from solutions that focus on creating agents and making them available to a team. In Tero, an agent isn’t simply a reusable prompt or shared configuration. It’s a specialized capability embedded in a QA workflow, with defined context, tools, controls, permissions, and outcomes.

Tero enables functional testers, test automation specialists, performance specialists, business analysts, QA leads, developers, product teams, managers, and other roles involved in quality decisions to share context, knowledge, and evidence, and use that information to better understand the systems they work with.

Ultimately, beyond connecting agents with tools, Tero controls which context, tools, and actions are available to each agent based on its QA workflow.

What Does Tero Offer Beyond Agent Building and Sharing?¶

Tero within Abstracta Intelligence¶

Tero is part of Abstracta Intelligence, Abstracta’s solution for bringing AI into software quality across the development lifecycle. Abstracta Intelligence builds on nearly 20 years of Abstracta’s experience in software quality and combines three components that continue to evolve month by month:

1. Tero Enterprise Edition: the agent harness and specialized agents
2. AIX — Abstracta AI Experience: AI adoption supported by Abstracta Academy
3. Impact Dashboards: measurement of results and value generated

1. Tero Enterprise Edition¶

Tero Enterprise Edition is the enterprise version of the open-source Tero project, backed by Abstracta. It enables organizations with strict control and regulatory requirements, such as banks, insurance companies, and government entities, to build and deploy agents securely.

It currently includes more than 40 ready-to-use agents for functional testers, automation engineers, performance specialists, business analysts, and managers, along with a growing community that shares knowledge, contributes new agents, and extends Tero through new connectors.

Tero integrates with the tools quality teams already use, including Jira, Confluence, PractiTest, YouTrack, Playwright, Selenium, Postman, GitHub, GitLab, Datadog, and SQL databases, among others.

It also integrates with key platforms in the financial sector, such as Bantotal, a core banking platform that processes financial activity for more than 50 million people, and IC Banking by Infocorp, a digital banking suite used by more than 40 banks. Its roadmap also includes new integrations with platforms such as PayStudio, a credit card management system.

We cover Tero’s integrations in more detail later in this article.

Other relevant capabilities include:

• Agent governance, evaluation, and testing. Each agent can include a test suite, budget controls, and limits on the number of steps it can execute, making it possible to evaluate performance consistently and establish a baseline for comparison when switching LLM providers.
• Observability and Chain of Thought. Tero makes it possible to inspect agent execution and the available reasoning trace to audit decisions, optimize prompts, and evaluate which LLM is the most cost-efficient for each task. This helps reduce the black-box effect and helps users move from consuming AI to creating and evaluating their own agents.
• LLM and infrastructure independence. Tero is model- and cloud-agnostic. It supports proprietary models such as GPT, Claude, and Gemini through APIs, as well as open-weight or local models such as Qwen, DeepSeek, and Kimi, deployed on-premise or through Bring Your Own Cloud (BYOC), reducing vendor lock-in.
• Open and auditable code. Organizations can inspect the code manually or automatically using their own security mechanisms. Because the codebase is open, organizations retain control over their workflows, context, and agent logic.
• Designed for QA and functional analysis. Tero provides a low-code interface for creating, testing, and maintaining agents without complex programming. It can be used directly in the workspace through Browser Copilot—from Jira, the application under test, or the web—and through MCP from tools such as Cursor, Claude Desktop, Claude Code, and GitHub Copilot. It also provides simplified access through web-based SSO, with no local installation required for end users.

2. AIX — Abstracta AI Experience¶

AIX provides the knowledge foundation teams need to identify opportunities, propose solutions, and implement AI agents effectively. It includes training, workshops, 1:1 working sessions, and ongoing support, backed by Abstracta Academy, so AI adoption does not remain limited to isolated efforts and instead becomes a sustained capability across teams and the organization.

3. Impact Dashboards¶

Measuring AI impact and return on investment is critical to sustaining these efforts. That’s why a Tero deployment includes a dedicated Impact Dashboards component: agents collect data and indicators from day one to track adoption and investment in AI. The standard dashboards are:

• Integrated adoption dashboard — AI Console: agent adoption metrics, time saved by agent and by person, sessions, and chats
• Quality practice maturity level: project score based on implemented practices, from basic testing to continuous testing, using Abstracta’s maturity model based on ISO 25010
• DevOps DORA Metrics: measurement of delivery speed and stability
• Release Readiness Report: consolidated indicator for evaluating whether a release is ready to move to production

The solution also supports custom Impact Dashboards tailored to each organization’s needs and objectives.

What Tools and Platforms Does Tero Integrate With?¶

As an agent harness specialized in QA, Tero has an open architecture that enables integration with the tools and context sources used throughout the software development and quality lifecycle.

This openness is important because quality work doesn’t happen in a single tool. It spans management systems, code repositories, test environments, and production monitoring. An agent that can access only part of that ecosystem ends up reasoning from incomplete information.

Tero currently integrates with:

• Management and collaboration: Jira, Confluence, Redmine, YouTrack, and Azure DevOps
• Test management: PractiTest and Zephyr
• Test automation and execution: Playwright, Selenium, Postman, and databases
• Repositories and version control: GitHub, GitLab, Azure DevOps, and repositories through the Codebase integration
• Continuous integration and delivery (CI/CD): Azure Pipelines
• Databases: SQL Server, MySQL, PostgreSQL, and other engines compatible with standard SQL
• Legacy systems and IBM platforms: access to and analysis of code and context in technologies such as COBOL and RPG, with support for IBM i / AS/400 environments
• Observability and monitoring: Datadog
• Knowledge and documentation: GeneXus Knowledge Base (KB), corporate documentation (Docs), and other document sources
• Web navigation and interaction: Browser, for automated interaction with web applications and evidence capture
• External information sources: Web, for accessing and extracting public information, and MCP (Model Context Protocol) servers for connecting with additional enterprise tools

This architecture allows Tero to integrate with the organization’s existing technology ecosystem and gives each agent access to the context, tools, and actions required to execute its QA workflow. New integrations can also be added when a project requires them.

How Are Tero Integrations Implemented?¶

The integration approach depends on each organization’s technology ecosystem, standards, and security requirements. Tero is designed to integrate with the existing architecture, minimize adoption effort, and define agent access to systems and actions according to the needs of each workflow.

Depending on the scenario, integrations can be implemented through:

• REST, SOAP, or GraphQL APIs, when applications expose integration interfaces
• SDKs or native connectors for tools such as Jira, GitHub, GitLab, Azure DevOps, or PractiTest
• Direct database access, when information needs to be queried or validated
• Integration with CI/CD pipelines, with Tero incorporated as another stage in the development and deployment process
• MCP (Model Context Protocol), for standardized connections with tools, repositories, and services that already provide MCP servers
• Integration with other agents or AI platforms, working with assistants already used within the organization through APIs, messaging protocols, or multi-agent architectures

When no standard integration mechanism exists, custom adapters can be developed to connect proprietary applications or legacy systems.

One example is the development of an MCP server for IBM AS/400 (IBM i), which allows AI agents to interact directly with applications and data in the AS/400 environment through a standardized protocol, without depending on command-line interfaces (CLI). This MCP can be deployed securely within the organization’s local infrastructure alongside the Tero installation.

What Tools Does Tero Use for Each Type of Testing?¶

Tero doesn’t depend on a single tool or framework. As an agent harness, it integrates with the tools each organization already uses to manage requirements, testing, execution, and defects, and adapts the technology selection to the type of test, the client’s stack, and its existing ecosystem.

Functional Testing¶

Tero can support different stages of the functional testing workflow, from requirements analysis and test design to execution, defect management, and results analysis.

Tero integrates with tools such as Jira, Azure DevOps, PractiTest, and Zephyr, where teams manage requirements, test cases, execution, defects, and evidence. It can also interact with web applications through Browser Copilot. When test cases are suitable for automation, Tero can connect with frameworks such as Playwright and Selenium.

Its capabilities include:

• Generating functional test cases from user stories, requirements, or acceptance criteria
• Identifying positive, negative, alternative, and edge-case scenarios that may not have been considered
• Reviewing functional coverage and detecting potential gaps in test cases
• Generating appropriate test data for each scenario
• Assisting testers during manual execution with steps and expected results
• Automatically generating execution evidence and documentation
• Creating defects with relevant information—steps to reproduce, impact, evidence, and context—integrated with tools such as Jira or Azure DevOps
• Recommending which cases are candidates for automation based on execution frequency, criticality, or stability
• Observability in pre-production environments: analyzing logs, metrics, traces, and events to detect anomalies, identify the root cause of failures, and correlate issues observed during functional testing with application and infrastructure behavior

Automated Testing¶

Tero supports Playwright as a modern standard for web/UI automation, but it’s framework-agnostic. It can work with Selenium, Appium, or custom frameworks executed through CI/CD. The tool is selected based on the type of testing, the client’s technology, and its existing ecosystem.

This flexibility allows Tero to adapt to each organization’s existing ecosystem, use the tools, frameworks, and processes the client already has in place, and add new integrations when needed. This accelerates adoption without replacing previous investments in automation or unnecessarily changing existing development and quality processes.

Performance Testing¶

Tero includes agents that assist in creating load-testing scripts in JMeter, Groovy—the language used within JMeter—and k6. These agents can be extended to work with other tools depending on the needs of the QA workflow and the organization’s technology ecosystem.

In projects where the client doesn’t already have a defined or licensed solution, Abstracta typically recommends JMeter DSL, a library for designing and executing performance tests as code (Test-as-Code) with the full capabilities of JMeter.

Web Services Testing (REST/SOAP APIs)¶

For API validation, Tero can chain agents that generate documentation from Swagger and code, then create scripts in specialized tools such as JMeter, REST Assured, or Postman/Newman. It also supports validation through Playwright API testing. Tool selection depends on the client’s technology, existing stack, and the type of test being automated. Tero can integrate the most appropriate tools for each QA workflow and work with the technologies the organization already uses.

What Tasks Does Tero Support Beyond Requirements and Testing?¶

Tero started with a focus on software quality and testing, but its AI agent-based architecture allows it to provide support across the entire software development lifecycle (SDLC). This makes sense because system quality is not defined only during the testing stage. It is also shaped by how a requirement is specified, how an API is documented, or how a backlog is prioritized.

Tero can currently support:

• Requirements analysis and definition: creating and refining user stories, acceptance criteria, and use cases
• Technical and functional documentation generation: specifications, API documentation, technical manuals, and architecture documentation
• Code analysis: code review, defect identification, code smells, vulnerabilities, refactoring opportunities, and explanation of legacy code
• Testing and quality: test case generation, automation, coverage analysis, execution and result analysis, defect investigation, regression analysis, risk assessment, and release readiness
• Architecture and solution design: support for architecture decisions, pattern definition, design reviews, and evaluation of technical alternatives
• Project management: support for backlog refinement, story estimation, prioritization, risk identification, and progress reporting
• DevOps and CI/CD: generation and review of pipelines, configuration validation, and support for continuous integration and deployment processes
• Knowledge management: querying corporate documentation, development standards, code repositories, and knowledge bases to answer technical questions with project context
• Metrics and analysis: integration with Apache DevLake to build customized software lifecycle dashboards

These agents work within the client’s existing ecosystem—including repositories, ALM tools, documentation, pipelines, and management systems—and act as specialized integrators for each stage of the development lifecycle, without requiring changes to processes or tools the organization has already adopted.

Real Tero Use Cases¶

These are some use cases implemented for current clients in Latin America and the United States. They show how Tero combines specialized agents, context, and tools across different QA workflows:

• Agent chain for API testing. Agents document the API from code or Swagger, design the test cases, then generate and run the test with JMeter.
• Legacy/mainframe system migration (for example, DB2/COBOL). Specialized agents analyze relational data schemas, generate data transformation and validation scripts, assist with end-to-end (E2E) testing, accelerate validation, and reduce bottlenecks without requiring constant manual intervention.
• Playwright automation agent. Starting from a browser session, it generates a test covering every API called by the frontend to validate the feature’s logic end to end.
• Observability for users of the Bantotal banking core**.** It helps testers and analysts better understand how Bantotal works internally, directly from the system, with access to the context they need to analyze its behavior. View the product in the Bantotal Store.

Together, these cases show how Tero applies specialized agents across different QA workflows, connects them with the context and tools of each environment, and supports consistent operation across modern and legacy systems.