OpenAI announced three simultaneous moves on June 11 that collectively accelerate its enterprise positioning
Executive Summary
OpenAI's acquisition of Ona gives enterprises persistent, auditable agent infrastructure that turns complex marketing workflows — budget pacing, cross-channel orchestration, creative sequencing — from engineering projects into delegatable processes. BBVA's 100,000-seat ChatGPT Enterprise deployment is the social proof that accelerates enterprise procurement globally. The window to establish internal workflow architecture before lock-in normalises closes around Q2 2027. Deploy a scoped agent on campaign budget reallocation by September 2026.
The Signal
OpenAI announced three simultaneous moves on June 11 that collectively accelerate its enterprise positioning. First, it confirmed plans to acquire Ona, a company that provides secure, persistent cloud environments, explicitly to extend Codex with the infrastructure needed to run long-duration autonomous agents across enterprise workflows. Second, BBVA disclosed it has deployed ChatGPT Enterprise to all 100,000 of its employees, making it one of the largest single-organisation AI deployments on record. Third, Wired profiled Thibault Sottiaux, the OpenAI engineer now leading a broad architectural overhaul of ChatGPT itself — the same figure who built out OpenAI's coding business into one of its fastest-growing revenue lines.
What Changed
Enterprises can now run persistent, multi-step AI agents inside secure, auditable cloud environments without building that infrastructure themselves. The Ona acquisition means Codex moves from a code-generation tool into a platform for long-running autonomous workflows — tasks that previously required human handoffs or custom engineering can now be delegated end-to-end to an agent that retains context and operates continuously across enterprise systems.
Why It Matters
The Ona acquisition is the move that matters most here, and its sharpest implication is this: OpenAI is building the infrastructure layer that makes enterprise AI irreversible. Once a 100,000-person organisation like BBVA runs persistent agents across its core workflows — not chatbot queries, but continuous, context-retaining processes — the switching cost becomes existential. You cannot rip out the nervous system of an operating bank. For enterprises running complex, multi-step marketing operations, this changes the calculus on what can be delegated. Campaign orchestration, budget pacing decisions, creative variant testing at scale, cross-channel sequencing — these are not simple prompt-response tasks. They require sustained context, handoffs across systems, and the ability to act without a human in the loop at each step. Persistent agent infrastructure makes all of that tractable for the first time without armies of engineers to build the scaffolding. What becomes obsolete is the integration middleware layer — the category of martech that exists primarily to stitch systems together and hand work between them. If an agent can maintain context and operate continuously across Salesforce, a DSP, a CDP, and a creative production system, the connective tissue tools lose their reason to exist. The "why behind the why" is competitive moat construction. OpenAI is not selling software — it is selling lock-in disguised as productivity. Sottiaux's architectural overhaul of ChatGPT, combined with the Ona infrastructure play, signals that OpenAI intends to own the enterprise workflow layer before any rival establishes a comparable footprint. The BBVA announcement is the proof point designed to accelerate that dynamic. At this scale, social proof becomes a closing argument.
Marketing Impact
martech
Integration middleware — the iPaaS connectors, workflow automation layers, and point-to-point stitching tools that sit between CDPs, DSPs, and CRMs — loses its primary justification. Persistent agents that retain context across systems make the connective tissue category redundant before most martech stacks have finished their current three-year contracts.
marketing ops
Campaign orchestration, budget pacing, and cross-channel sequencing become delegable to continuous agents rather than managed through human-supervised workflows. The marketing ops function shifts from process management to agent governance — defining rules, auditing decisions, and setting guardrails rather than executing handoffs between platforms.
product marketing
Competitive intelligence and launch readiness workflows — historically bottlenecked by research synthesis and cross-functional coordination — can now run as persistent background processes. Product marketers who redeploy that capacity toward positioning judgment and narrative development will outpace those still managing the logistics of information gathering.
The Exploit
Opportunity
Marketing operations leaders can now pilot end-to-end autonomous campaign workflows — budget pacing, creative variant sequencing, cross-channel scheduling — inside ChatGPT Enterprise without building custom agent infrastructure. Teams running Salesforce, a DSP, and a CDP in parallel can delegate the connective logic to a persistent agent, cutting integration overhead and reducing campaign-ops headcount costs by an estimated 20-30% within 12 months of deployment.
Risk
Deploying persistent agents across live campaign infrastructure before enterprise audit and rollback standards mature exposes teams to compounding errors at scale — a misfire in budget pacing is not a prompt to retry, it is real spend gone.
The Move
By September 2026, have your marketing operations lead deploy a scoped ChatGPT Enterprise agent across one high-volume, bounded workflow — campaign budget reallocation is the right first candidate — with audit logging enabled and a defined human-override trigger. Success checkpoint: agent handles 80% of reallocation decisions without escalation within 60 days.
First-Mover Advantage
Gains
First movers establish internal workflow architecture around persistent agents before vendor lock-in normalises — capturing productivity gains while retaining negotiating leverage over OpenAI on pricing and data governance terms.
Risks
Deploying persistent agents across live campaign infrastructure before enterprise audit and rollback standards mature exposes teams to compounding errors at scale — a misfire in budget pacing is not a prompt to retry, it is real spend gone.
Window
The advantage window runs roughly until Q2 2027, when rival platforms — Google Agentspace, Salesforce Agentforce — reach comparable persistent-context depth. The signal that closes it is cross-platform agent interoperability becoming a default procurement expectation.
Winners & Losers
Winners↑
Enterprise marketing operations teams
Persistent agent infrastructure converts multi-step marketing workflows — campaign orchestration, budget pacing, cross-channel sequencing — from engineering projects into delegatable processes, dramatically compressing execution timelines without proportional headcount increases. The mechanism is context retention across sessions and systems, which eliminates the human handoff bottleneck that has historically made full automation of complex workflows impractical. Teams should begin mapping their highest-friction, multi-system workflows now and identify which are structurally suited for persistent agent delegation before competitors do.
OpenAI enterprise distribution partners
The BBVA deployment at 100,000 employees functions as a closing argument for every undecided enterprise still in pilot mode — social proof at this scale shifts the procurement conversation from 'should we' to 'how fast.' Partners with existing enterprise relationships and implementation capability are positioned to capture accelerated deal flow as OpenAI's enterprise credibility compounds. The action is to front-load case study development and executive briefings in Q3 2026 before this window becomes a commodity.
In-house marketing technology teams
Organisations with strong internal martech capability gain disproportionate advantage from persistent agent infrastructure because they can integrate it directly into proprietary data environments without dependency on third-party middleware to broker connections. The mechanism is that Codex-as-agent-platform rewards teams that already understand their own system architecture — external agencies and generalist consultants cannot move as fast. The play is to accelerate internal Codex and OpenAI API experimentation now, before the workflow patterns that deliver outsized ROI become established and widely imitated.
AI-native creative production platforms
Persistent agents that can operate continuously across enterprise systems — including creative production environments — create native integration opportunities for platforms already structured around API-first, modular workflows. The mechanism is that long-running agents need creative systems that can receive instructions, execute variants, and return outputs without human-mediated handoffs, which structurally advantages platforms built for programmatic creative over legacy tools designed for human-in-the-loop production. These platforms should be actively building OpenAI agent compatibility documentation and positioning for enterprise procurement conversations happening in Q3 2026.
Losers↓
Martech middleware and integration platform vendors
The explicit value proposition of middleware — connecting disparate enterprise systems and passing work between them — is the precise function that persistent AI agents are designed to absorb. Once an agent maintains context across Salesforce, a DSP, a CDP, and a creative system without a connector layer brokering each handoff, the commercial rationale for standalone integration tooling collapses. Vendors in this category should accelerate their pivot toward agent orchestration and governance tooling, because the connective tissue play has a shrinking runway.
Marketing systems integration consultancies
Professional services firms whose billable work centres on architecting and maintaining integrations between marketing technology platforms face the same structural threat as the middleware vendors they configure — persistent agents reduce the complexity that justifies the engagement. The mechanism is displacement from below: tasks that previously required scoped implementation projects become configuration decisions within an agent platform. The defensive move is repositioning toward agent governance, prompt engineering at enterprise scale, and workflow audit services — but the window to establish that credibility before the market commoditises it is narrow.
Competing enterprise AI platform vendors
The combination of Ona's persistent cloud infrastructure, the BBVA proof point at 100,000-seat scale, and Sottiaux's architectural overhaul of ChatGPT simultaneously raises the capability floor, the switching cost floor, and the social proof bar that any rival must clear to compete for enterprise marketing workflow contracts. The mechanism is compounding lock-in: once persistent agents are embedded in core marketing operations, displacement requires not just a better product but an organisational willingness to re-engineer live workflows under production conditions. Rivals need a credible persistent agent infrastructure story and a marquee reference customer before Q4 2026 or cede the enterprise workflow layer entirely.
Strategic Outlook
The pattern from the past eighteen months of enterprise SaaS consolidation repeats here at higher velocity: the platform that embeds at workflow depth first wins the renewal cycle, not the one with the best feature set. OpenAI is executing the same playbook Salesforce ran in the 2010s — anchor to a high-frequency use case, expand surface area, make displacement economically irrational. BBVA's 100,000-seat deployment is the lighthouse reference that moves procurement conversations at every competing financial institution and large enterprise globally. Expect Google DeepMind and Anthropic to accelerate their own persistent agent infrastructure announcements in Q3 2026, and expect Microsoft to respond by deepening Copilot's integration with Azure's secure compute layer. The martech vendors most exposed are those whose entire value proposition is orchestration and connectivity — they have roughly two to three product cycles to pivot toward agent-native tooling or face structural irrelevance by 2027.