G2

Verified User in Accounting

Small-Business (50 or fewer emp.)

9/3/2026

"Flexible goal-based agents with a strong Python ADK and IBM automation integration"

4/5

Describe the project or task IBM watsonx Orchestrate helped with:

The service task involves setting up and managing flexible goal-based agents using the Python ADK and integrating them with IBM's automation tools. This includes defining agents by their goals and tool inventories, rather than traditional flowcharts, allowing for dynamic and adaptable processes.

Key tasks include:

1. Initial setup of the environment and custom tools.

2. Registering custom tools using decorators and docstrings.

3. Selecting appropriate models from the watsonx.ai catalogue.

4. Integrating with IBM RPA for systems without APIs.

5. Debugging and testing agent behavior.

6. Enhancing reasoning logs for audit purposes.

The goal is to leverage agentic automation to streamline processes without requiring a modern system landscape. Review collected by and hosted on G2.com.

What do you like best about IBM watsonx Orchestrate?

The thing that has held up across every project is that Orchestrate lets me define an agent by its goal and its tool inventory rather than by a flowchart. That sounds academic until you hit a process where the branches genuinely can't be enumerated in advance — then it's the difference between a project that ships and one that doesn't.

The Python ADK is the part I use most. Registering a custom tool is a decorator and a docstring, and the docstring is what the agent actually reasons over, so the boundary between code and agent behaviour stays readable. Being able to pick the underlying model from the watsonx.ai catalogue matters too — I run Granite for German-language work where governance and data residency are non-negotiable.

The other thing I value is how well it plays with the rest of the IBM automation stack. Being able to fall back to IBM RPA when a target system has no API has rescued more than one project. Review collected by and hosted on G2.com.

What do you dislike about IBM watsonx Orchestrate?

The initial setup is the weakest part of the experience. Getting from an empty environment to a first working custom tool involves more moving parts than it should, and the documentation assumes you already know which of them you need. My first project lost days to that, and it's the main thing I warn clients about when we scope.

Debugging agent behaviour is still harder than debugging a workflow. When an agent chooses the wrong tool or escalates when it shouldn't, the fix is usually in the system prompt rather than in code, and there's no good way to test that systematically — you iterate by observation. Better tooling for evaluating and regression-testing agent decisions would be the single most valuable addition.

Reasoning logs are readable but not yet audit-grade out of the box. In regulated environments I end up building my own logging layer, and that's effort every customer pays for separately. Review collected by and hosted on G2.com.

Recommendations to others considering IBM watsonx Orchestrate:

To improve the initial setup experience, consider providing a more guided setup process with clear step-by-step instructions and a checklist of necessary components. This could help new users avoid the pitfalls of missing parts and reduce the time spent on initial configuration.

For debugging, developing a more robust testing framework that allows for systematic evaluation of agent decisions could greatly enhance the user experience. This framework could include tools for simulating different scenarios and logging agent responses for analysis.

Enhancing the audit capabilities of reasoning logs would be beneficial, especially in regulated environments. Providing an out-of-the-box solution for audit-grade logging could save users the effort of building custom solutions and ensure compliance with industry standards. Review collected by and hosted on G2.com.

What problems is IBM watsonx Orchestrate solving and how is that benefiting you?

Two live examples from my work. In property management, tenants report damage via WhatsApp in free-form language, usually incomplete. Orchestrate interprets the report, asks back for what's missing, classifies it, and produces a structured case — which then gets written into an ERP so old it exposes no API at all, via IBM RPA driving the UI. The benefit is that a channel people actually use now connects to a back office that was never designed for it, without an ERP migration first.

The second is procurement. Long-tail spend is where price increases slip through because renegotiating a small amount costs more effort than it saves. An Orchestrate-based negotiation agent argues from market indices, applies a defined tolerance corridor, and escalates to a human buyer above configurable thresholds. It closes the gap between what's worth negotiating manually and what's worth negotiating at all.

In both cases the real win is the same: agentic automation doesn't require a modern system landscape. Review collected by and hosted on G2.com.