Contents
Clay has made auto-mode the default setting for Claygent, its AI research agent, so each row in a table is now routed to whichever AI model suits that task rather than one model chosen up front. Clay announced the change on 27 October 2026 and said that on production traffic, auto-mode cost 47% less than sending every task to the most capable model, "while maintaining frontier quality".
For GTM engineers, the practical change is that the cost of an AI research column is no longer fixed by a single setting. It now depends on how hard each row is.
What did Clay launch?
Until now, a Claygent column ran every row on the model picked when the column was set up. Auto-mode replaces that with a router that decides per row. Clay says simple lookups go to fast, cheaper models and complex research goes to the most capable ones, and each routing decision takes under a second.
The details from Clay's announcement:
1/ Auto-mode is the default for new Claygent columns in tables, new nodes in Clay Workflows, and when testing in Claygent Builder.
2/ Existing Claygents keep the model they use today, as do duplicates of them. Any column can still be switched to a specific model.
3/ There is no charge for auto-mode itself. Users pay for whichever model runs each task.
4/ Clay says auto-mode enforces a quality floor for every task category, so it cannot save credits overall by doing worse on one kind of research, such as contact lookups.
The router was built with Applied Compute. Clay says it post-trained an open-weight model with the company to act as the router, which runs on Applied Compute's platform and handles "hundreds of millions of queries a month".
Why does per-row routing matter?
Because rows in a prospect list are not equally hard. Clay's own example: checking whether a well-known public company sells to businesses usually takes one search, but the same question about a small startup with a bare website can take several searches and more careful reading.
Under a single-model setup, a team had two bad options. Pay frontier prices on the easy rows, or send a weaker model to the hard ones and accept worse answers. "Picking the strongest model to be safe meant paying for it on every row," Clay wrote.
What does 47% look like on a real budget?
Take an illustrative team spending £2,000 a month on AI research credits, all of it on the most capable model. If its workload behaved like Clay's production traffic, the same work under auto-mode would cost about £1,060 a month, a saving of roughly £940, or around £11,300 a year.
That is an average, not a promise. A list made mostly of small, poorly documented companies will lean towards the expensive models, and the saving will be smaller. Clay's own advice is to test auto-mode on a sample of five to ten rows before running a large table.
What should GTM engineers do now?
Three things follow from the launch.
1/ Re-test existing workflows. Old Claygent columns do not switch over automatically. If a column runs a high-volume, simple task on a frontier model, it is a candidate for moving to auto-mode, but compare outputs on a sample first.
2/ Keep fixed models where consistency matters. Some workflows need the same model on every row so results are comparable over time, such as a scoring prompt feeding a routing rule. Clay lets you pin a model for exactly this reason.
3/ Spend the savings on the inputs. A cheaper research step does not fix a wrong record. Clay itself reported on 21 October 2026 that a nightly review of one of its own lead routing workflows found 47 of 159 leads placed in a segment that did not match their company size. Model routing lowers the cost of each answer. It does not check whether the account behind the row was the right one to research.
Where does this fit in Clay's year?
The launch follows Clay's $115 million Series D at a $7.1 billion valuation, announced on 9 September 2026, when the company said it had more than 17,000 customers. Since then it has shipped agent tooling at a steady pace, and cost control is becoming part of that pitch: as AI research moves from occasional columns to always-on workflows, the bill per row starts to matter as much as the quality of the answer.
A note on Peach Data
Peach Data builds go-to-market engineering systems for sales and marketing teams: the data foundation, the workflows and the agents behind outbound and inbound. We started as a verified data business, so every build starts with the data layer. peachdata.co.uk
