OpenAI has unveiled a new real-world implementation of self-improving AI agents built using Codex, demonstrating how autonomous AI systems are beginning to reshape complex professional services workflows such as tax preparation and accounting. The project was done in partnership with Thrive Holdings and Crete Professionals Alliance. It shows the way agentic AI systems update themselves permanently by learning from operational feedback loops instead of relying on the one-time fish training ones only.
Per OpenAI, this system employs AI Codex-powered agents to support accountants in the tax preparation process by creating tax returns, checking tax documents, finding errors, and improving results via continuous iterations. It fuses the knowledge of the professionals with the automatic feedback loops of the AI system giving the latter the ability to learn and increase its performance over time.
The project reportedly achieved significant operational gains during pilot deployments across accounting firms within the Crete network. OpenAI and Thrive stated that the system improved drafted tax return accuracy from roughly 25% at launch to as high as 97% within weeks by using real-world corrections and workflow data as learning signals. The deployment also reduced tax preparation time by approximately one-third while increasing operational throughput.
The announcement is reflective of the industry-wide trend towards “self-improving” AI agents that are able to improve their own workflow through repeated use and feedback. This technology, rather than merely acting as passive assistants, will be developed in such a way as to perform tasks, assess outcomes, and continuously improve efficiency.
Also Read: Snowflake Expands AWS Partnership With $6 Billion Commitment to Accelerate Enterprise Agentic AI Adoption
Agentic AI Moving Beyond Productivity Tools
This launch highlights the fact that enterprise AI is moving beyond basic automation and chatbot features into fully operationally autonomous systems that can be used to manage advanced business processes. The initial implementations of agentic AI systems were geared mainly towards generating content, assisting in searching, and increasing productivity. But the most recent version of agentic AI systems is designed to be used to automate multistep processes and interact with other software tools.
The Codex AI tool developed by OpenAI has emerged as a major driver behind this transformation. The Codex is an AI system that was designed primarily for coding and coding assistance but it is increasingly being leveraged as a more general enterprise AI system for working with data pipelines, application program interfaces (APIs), and workflow tools.
The tax preparation use case is particularly significant because accounting and tax workflows involve high levels of complexity, regulatory interpretation, documentation management, and domain-specific reasoning. Successfully deploying self-improving AI agents in such environments could accelerate AI adoption across other professional service industries including legal services, consulting, insurance, and financial operations.
Impact on the IT Industry
For the IT industry, this development underscores the growing significance of the AI agent infrastructure, workflow orchestration, and enterprise automation platforms. Firms are now in need of technology that would enable the deployment of AI systems within their operational environment and guarantee scalability, governance, security, and compliance.
Self-improving AI agents could also give rise to further spending on enterprise AI engineering, feedback loop architecture, retrieval systems, observability platforms, and governance solutions for AI technology. Companies using autonomous AI agents will certainly need infrastructure that enables them to monitor performance, verify outputs, manage permissions, and guarantee compliance in critical environments.
Finally, the project emphasizes the growing convergence of software engineering and operational AI technology. Increasingly, AI agents are viewed as an operational technology that continually evolves through interacting with business processes and improving them in this way.
Business Impact and Future Outlook
Self-improving AI entities can revolutionize operational efficiency and productivity in many businesses. Service-oriented businesses, for instance, are known to struggle with recurring workloads, staff shortages, and higher demands from customers. Self-learning AI entities can enable businesses to expand their capacity without having to increase staffing levels at the same pace.
On the other hand, highly skilled employees will be able to take advantage of self-improving AI entities to free up time and enable them to undertake valuable strategic advisory roles. Such improvements are especially important in highly specialized industries like accounting and tax preparation.
It should also be noted that self-improving AI entities are also likely to spark conversations about transparency, accountability, and auditability in business. The increased deployment of AI entities implies the need for more effective frameworks to ensure that outputs are accurate and meet the necessary requirements.
OpenAI’s self-improving tax agents ultimately illustrate how enterprise AI is entering a new phase where systems are no longer limited to assisting workers but are beginning to actively participate in and optimize operational business processes. As agentic AI capabilities continue advancing, industries built around knowledge-intensive workflows may undergo significant transformation in the years ahead.