Meta says its Intelligent Delivery system has improved ad-ranking performance by about 10% by steering limited computing capacity towards the requests and workloads where it can do the most. The system also uses AI agents to help find and test optimisations, with engineers approving changes before they ship.
Meta AI Watch analysis
What happened
In a post published on 11 November, Meta Engineering describes Intelligent Delivery, a system that allocates resources across the company’s ad-serving operation. Meta says it has contributed an additional roughly 10% improvement in ad-ranking performance, on top of model and system advances made from 2017 through the first half of 2026.
The system uses machine-learning models to estimate which ad requests are likely to benefit from more compute, then coordinates decisions about data-centre regions, hardware and the amount of host capacity each request receives. Meta says one model improved its ability to identify the highest-impact requests by about 20% in offline benchmarks compared with its previous production model.
Meta also says AI agents contributed to about 20% of Intelligent Delivery optimisations in the first half of 2026. On its most automated mechanism, they now account for about 80% of the engineering effort, from analysing system changes to proposing and running experiments. Engineers review and approve the final changes. Meta says that process has shortened optimisation cycles from months to weeks.
Why it matters
The practical point is not simply that Meta has added AI to its advertising stack. It is trying to get more useful work from finite compute, while balancing ranking performance against latency and the different capabilities of its hardware. That matters as models grow and serving capacity remains a constraint.
Meta’s figures describe its own system and results, not an independent comparison. Still, the account offers a detailed example of AI agents being used inside a production engineering loop, with people retaining sign-off rather than handing over the keys.
Our read
This is a more grounded AI story than the usual promise that an agent will transform everything before lunch. Meta describes a specific job, a measurable claimed benefit and where engineers remain in control. The interesting test is whether shorter optimisation cycles translate into sustained improvements, not merely faster experimentation.
What to watch
- Whether Meta reports further performance or efficiency results from Intelligent Delivery.
- How much of the optimisation process agents take on beyond the mechanisms already automated.
- Whether human review remains effective as the system and its search space grow.
Discussion spark: Should AI agents be allowed to run most of an optimisation process when engineers still approve the final change, or should people remain more directly involved throughout?
Sources and evidence
- Intelligent Delivery: Making Every Unit of Compute Serve Better Ads (11 November 2026, 17:00 UTC)
not affiliated with, endorsed by, or operated by Meta