OpenAI tests agentic AIs
OpenAI ran an internal hackathon experimenting with autonomous, agent-based AIs to run finance workflows — moving beyond static models that only assist humans. Boards now expect rapid value from AI deals, but firms lack governance: 55% have hired AI roles while only 44% completed impact assessments, creating a real governance gap. (edtechinnovationhub.com) (stocktitan.net)
An internal finance hackathon at OpenAI engaged more than 200 employees to build and deploy Codex-powered agentic workflows directly into day-to-day finance operations, according to CFO Sarah Friar’s LinkedIn post. (edtechinnovationhub.com) Teams delivered end-to-end prototypes that included automating audit outputs into structured formats, using diligence agents to analyze M&A data rooms, and converting contracts and invoices into standardized statements of work. (edtechinnovationhub.com) Friar framed the effort as a shift “from tools that assist, to agents that execute,” and wrote that agents will be treated as organizational resources with employees accountable for managing them. (edtechinnovationhub.com) OpenAI has also published technical detail on an internal data agent (January 29, 2026) built with Codex, GPT‑5, the Evals API, and the Embeddings API that serves roughly 3.5k internal users and reasons across 70k datasets. (openai.com) A Gallagher survey of more than 1,200 global businesses (published March 24, 2026) found 62% have delivered AI training to employees while 55% have hired AI-focused roles, highlighting rapid skills investment alongside vendor experimentation. (stocktitan.net) The same Gallagher survey reported only 44% of organizations have conducted AI impact assessments and about 43% have yet to introduce formal AI risk-management frameworks, signaling a material governance shortfall as firms operationalize agents. (stocktitan.net) The OpenAI post (March 25, 2026) and Gallagher findings together document a near-term operational push inside a major AI lab and a concurrent industry-wide gap: rapid agent deployment at scale inside teams versus incomplete organizational risk and impact controls. (edtechinnovationhub.com)
