Tenex Sold AI Strategy. Now It Is Selling Agents.

Alex Lieberman announced Tenex Agents on September 22, describing them as “off-the-shelf, enterprise-grade AI agents” for recurring work across functions. That is a product launch. It also points at a harder change in the services business.

Tenex has sold AI transformation work: audits, adoption plans, change management, and engineering squads. Now it says some of that work can live inside a reusable agent. For a founder-marketer, that changes the question to ask an outside partner. Do you leave with a deck and a recommendation, or with a system that keeps doing a defined job after the engagement ends?

The agent is only the visible part

Tenex’s current transformation offer starts with 30-, 60-, or 90-day audits. They identify use cases, quantify ROI, and map an implementation plan. The company says each engagement ends in an AI Adoption Report and technical appendix that a customer can take forward itself or hand back to Tenex for execution.

Its engineering offer uses a similar framing. Tenex sells subscription-style engineering squads and says customers pay for delivered features rather than hours. The company’s public playbooks also turn pieces of its approach into guides, including competitive-intelligence research, AI evaluation, customer-health monitoring, and search strategy.

Tenex Agents appears to be the next packaging step. The announcement doesn’t list the agents, their integrations, pricing, customers, or performance results. So the story isn’t that Tenex has proven a new category. The story is that an AI services firm is testing whether its repeatable know-how can become ongoing software.

That matters to marketers because the same decision is coming to every specialist partner. A strategist’s useful work is a set of recurring judgments: which customer themes deserve a campaign, which claims need review, and which competitor shift should change positioning. It also includes which content idea has enough evidence to publish. Those judgments usually arrive as a workshop, a report, or a recurring call.

An agent can make a defined version of that work available more often and to more people. It can also expose how much of the value was a repeatable process and how much depended on the person doing the work.

The new buying test

Marketing teams shouldn’t buy an agent because it carries a partner’s brand. They should ask four plain questions before replacing a service deliverable with one.

What job does it own? “Marketing strategy” is a sales phrase. A useful agent has a bounded job, such as reviewing approved claims before a draft reaches legal or turning customer-call patterns into a weekly content brief.

What does it get to read and change? The answer needs to name the systems, data, permissions, and human approval points. A workflow without access to the right context becomes a better chat window. A workflow with uncontrolled access becomes a risk.

How will the team judge it? The service provider should be able to show the historical examples, rubric, and failure cases used to test the work. Those materials let a team tell a repeatable process from plausible output.

Who improves it after the first month? The useful asset includes a feedback loop that records bad decisions, updates the context, and assigns ownership for the next version. The maintenance work behind AI agents is usually where a promising demo becomes an operating cost.

Those are buyer questions, but they are also product questions for agencies and consultants. AI keeps lowering the cost of producing a slide, analysis, draft, or campaign variation. A partner with a distinctive method can now decide whether to keep billing for each instance of that work or codify part of it into a product.

A services model under pressure

The claim that AI can replace an agency is already common. We covered a group of startups selling that premise in July. Tenex starts from the partner’s own process rather than a generic AI growth-team promise.

Generic outputs are easy to copy. A useful agent needs the judgment rules, source material, edge cases, and quality bar that a good operator has built over time. Putting those things into a system is work. It takes more than renaming a prompt.

For a marketing leader, the near-term opportunity is smaller than the pitch. Don’t try to turn the whole function into an agent. Pick one recurring decision with good examples, a clear owner, and a measurable downside when it’s wrong. Build or buy the system around that job, then compare its work with the process it’s supposed to replace.

Lieberman’s announcement is early evidence that this packaging shift is reaching the firms advising companies on AI, not only the software vendors selling into them. Tenex has not shown enough public detail to prove the product works. But it has put a useful question on the table: when a partner’s best knowledge becomes software, what part of the relationship is still worth paying people to do?