Quick answerExisting customers generally have a reasonable expectation that an AI agent's documented commitments, refund policies, escalation guarantees, and prior representations survive an acquisition, even if the acquirer wants to change how the agent operates going forward. The safest path is to honor existing commitments for existing customers through a defined transition window, communicate any real changes clearly before they take effect, and treat undocumented but consistently-delivered behavior as an implicit commitment worth checking with legal before quietly changing.
This is a different moment than the due diligence that came before it
Evaluating a target company's AI agent before the deal closes, the process covered in AI Agent Due Diligence Before an Acquisition, is about assessing risk and value before you own the thing. This post picks up after the deal has closed, when the acquirer already owns the agent and its history, and now has to decide what changes and what stays the same for the customers who were already relying on it.
That is a materially different problem, because it involves real customers with real expectations formed under the previous owner, not a hypothetical risk assessment. A change that looked like a minor operational cleanup in the due diligence spreadsheet can land very differently once it actually happens to a customer who remembers being told something specific.
What counts as a "promise" worth honoring
Not every past behavior of an AI agent rises to the level of a commitment, but several categories generally do:
Explicit documented commitments: refund policies, SLA terms, escalation guarantees, or anything stated in a terms of service or contract that predates the acquisition. These are the clearest case, and generally the acquirer inherits them as a matter of contract law, not just goodwill.
Specific representations made directly to a customer, even informally, such as an agent telling a customer during a support conversation that a particular exception would be honored going forward. These are murkier legally, but customers experience them the same way as a formal commitment, and reneging on them without acknowledgment damages trust in a way that shows up in churn and complaints even if it never becomes a legal dispute.
Consistent behavior a customer has come to rely on, even without an explicit statement. If the agent has reliably offered a certain kind of flexibility for a year, customers build workflows around that reliability. Changing it silently, even if nothing was ever formally promised, tends to feel like a broken promise from the customer's side regardless of the legal technicality.
Building a transition plan instead of a silent cutover
Inventory what actually changes. Before announcing anything, get a precise list of what the acquirer intends to change about how the agent operates, distinguishing genuine behavior changes from internal-only changes like a provider migration that should be invisible to the customer, the kind covered in Migrating an AI Agent Between LLM Providers.
Sort changes by whether they affect existing commitments. A model swap that keeps behavior consistent is very different from a policy change that alters what customers can expect. Only the latter needs a customer facing communication plan.
Give existing customers a defined honoring window for commitments the acquirer plans to change going forward, rather than an abrupt cutoff. This mirrors the responsible approach described in A Responsible Customer-Facing Plan for Sunsetting an AI Agent: customers who built expectations under the old terms deserve notice and a transition period, not an instant change with no warning.
Communicate proactively, before customers discover the change themselves. A customer who finds out a commitment quietly disappeared, especially by having a request declined that would previously have been honored, loses much more trust than one who was told in advance, even if the underlying change is identical.
Where this intersects with a broader integration decision
If the acquirer is merging the acquired company's AI agent into its own, rather than running it as a standalone product, the mechanics of that integration are covered in How to Reconcile or Migrate Two AI Agents When Companies Merge. The commitments question sits above that mechanical integration: regardless of how the two agents technically get merged, the acquirer still has to decide, deliberately, which of the acquired agent's past promises to preserve during and after that merge, rather than letting the answer be whatever falls out of the technical integration path by default.
FAQ
Is the acquirer legally required to honor every past commitment? It depends heavily on what was formally documented in contracts versus what was informal behavior, and on the jurisdiction and deal structure. Get legal counsel involved specifically on customer facing commitments during deal integration planning, not just on the deal's core commercial terms.
How long should a transition window be for commitments the acquirer wants to change? Long enough that affected customers can reasonably adjust or make an alternative arrangement, which varies by the nature of the commitment. A refund policy change might need a shorter window than a long-term contractual guarantee.
Should the AI agent itself explain that a change is due to the acquisition? Generally yes, to the extent it is accurate and approved by legal and communications. Attributing a change to an unnamed policy update when customers already know an acquisition happened tends to read as evasive, which undermines trust more than a direct explanation would.