# Absorb — OpenAI Independent public example · September 2026 Independent Absorb assessment using public information only. Not affiliated with or endorsed by OpenAI. ## 1. Product summary Frontier-model access combined with agents, connected tools, enterprise controls and developer APIs. **User / buyer:** Individuals, developers and organizations; enterprise buyers include IT, operations and functional teams. **Core job:** Turn general AI capability into useful work—and increasingly into controlled multi-step execution. ## 2. Evidence and unknowns ### Observed facts - Workspace agents can run recurring workflows, use connected tools and operate with approval gates and role controls. [OpenAI — Workspace agents](https://openai.com/business/workspace-agents/) - Business plugins and MCP-based actions can read from and write to external systems under organizational controls. [OpenAI — Business plugins](https://openai.com/business/plugins/) - Enterprise usage is shifting from assistance toward delegated agentic work across business functions. [OpenAI — Enterprise Signals](https://openai.com/signals/enterprise-data/) ### Assumptions - Enterprise distribution and governance remain materially differentiated. - Frontier model quality still influences enterprise buying decisions. ### Critical unknowns - How much retention comes from model quality versus workflow and governance lock-in. - How quickly frontier intelligence becomes economically interchangeable. ## 3. Score, verdict and confidence 83.75/100 — 🟢 Grows with AI. Confidence: medium. This is not a probability of survival. ## 4. Eight weighted dimensions Weighted points = score / 4 × weight. ### Absorption Resistance Score 3/4 · Weight 20 · Points 15 Chat is copyable; governed actions, enterprise context and platform distribution are not. ### AI Progress Advantage Score 4/4 · Weight 15 · Points 15 Better AI expands the amount and complexity of work the platform can perform. ### Compounding Assets Score 3/4 · Weight 15 · Points 11.25 Shared agents, configurations, workflows and organizational adoption accumulate. ### Workflow Integration Score 3/4 · Weight 15 · Points 11.25 Schedules, tools and approvals place agents inside recurring work. ### Replication Difficulty Score 4/4 · Weight 10 · Points 10 Interfaces are easy to copy; models, infrastructure, distribution and governance are not. ### AI-Expanded Problem Score 4/4 · Weight 10 · Points 10 More delegated AI work creates more need for orchestration, context and control. ### Model Independence Score 1/4 · Weight 5 · Points 1.25 Model capability remains central rather than a commodity supplier. ### Defensive Path Score 4/4 · Weight 10 · Points 10 The product already extends into governed workflows and enterprise control. ## 5. Three reasons for the verdict - AI progress expands the workload OpenAI can address instead of removing demand. - Enterprise permissions, tools and approvals create value below the chat interface. - The structural weakness is model dependence if frontier intelligence commoditizes. ## 6. Biggest absorption risk Frontier intelligence commoditizes faster than OpenAI deepens ownership of workflow and control. ## 7. Strongest defensible asset today Enterprise distribution plus governed agent execution across connected tools. Asset existence does not establish causal retention or switching cost. ## 8. Survival Design Proposed future capabilities are not counted as existing strengths. ### Absorbable Shell Generic chat, summarization, drafting, search and simple assistants. ### Defensible Core Governed execution: permissions, approvals, auditability, shared agents, connected tools and developer distribution. ### Value Migration From “best answer” to trusted delegation across real systems with controls and recorded outcomes. ### Own the Source, Not the Interface Own the execution substrate, tool connections and policy layer that any AI interface still needs to call. ### Reality Anchor External actions, approvals, permissions, audit logs and workflow state. ### Workflow Evolution Define → connect context/tools → execute → approve → record outcome → improve → repeat. ### AI as a Channel Expose APIs and execution endpoints other AI experiences can invoke. ### Time to Moat A moat already exists; the next milestone is durable outcome history and repeatable agent workflows that are costly to unwind. ### Redesign Recommendation Keep moving from general AI destination to governed execution and control layer for organizational AI work. ## 9. Build decision Build and strengthen moat The current product benefits from AI progress; durable value should keep migrating from model/UI differentiation toward governed execution. ## 10. Decisive question If comparable frontier models were commodity-priced tomorrow, how much enterprise usage would remain because OpenAI owns the workflow and control layer? ## Seven-day validation plan Proposed experiment only. No research results claimed. Small-sample thresholds are discovery hypotheses, not proof of demand or defensibility. ### Days 1–2 **Do:** Observe five authorized business users running one repeated cross-tool task. Record permissions, approvals, current time and failure modes. **Ask:** What stops you from delegating this entire task today? **Measure:** Identify the same recurring task in at least three users and a measurable acceptance criterion. **Decide:** If tasks are unrelated, choose a narrower workflow before prototyping execution. **Owner:** **Observed evidence:** **Actual result:** **Next decision:** ### Days 3–5 **Do:** Run a supervised task rehearsal with existing approved tools. Log proposed actions, approval decisions, corrections and outcomes. Do not grant new permissions. **Ask:** Which actions need your approval, and what evidence makes you comfortable approving? **Measure:** Three users complete the task with correct state, explicit approval and no unauthorized action. **Decide:** If control or correctness fails, repair that specific boundary before expanding the task. **Owner:** **Observed evidence:** **Actual result:** **Next decision:** ### Days 6–7 **Do:** Compare a second run with the baseline. Interview the same users about which context, policy and history would remain useful if the underlying model changed. **Ask:** If model quality were equal across vendors, which part of this workflow would still make you stay? **Measure:** At least three users reuse the workflow and identify a concrete non-model dependency. This is exploratory evidence only. **Decide:** If passed, test repeat usage and operational switching cost for several weeks before inferring durable retention. **Owner:** **Observed evidence:** **Actual result:** **Next decision:** ## Method and limitations Absorb scoring guide. Fixed verdict bands: 80–100 Grows with AI; 60–79.99 Needs stronger defensibility; 40–59.99 At risk of absorption; 0–39.99 Likely to be absorbed. The build decision is reasoned separately. Dated public-information assessment. Not an audit or a forecast of market demand, revenue or financial performance.