Coforge has expanded its AgenticOps capabilities, pitching enterprises an operational foundation for running AI agents at scale. The announcement lists what breaks when agent fleets hit production: governance gaps, agent drift, new attack vectors, unreliable tool integration, data-sovereignty exposure, and escalating token costs. Agentic AI operations, the company argues, is what turns pilot autonomy into production autonomy.
The offering centers on EvolveOps.AI, described as the AgenticOps layer inside Coforge’s Nuuron suite. It provides continuous evaluations, automated drift detection, token-consumption monitoring, and a governed registry of approved tools and protocols. One harness, the pitch goes, replaces the fragmented tooling that strands most agent programs in pilot stage.
The numbers arrive with fanfare: more than 160 clients on the AgenticOps journey, 37 of them already seeing 98.2 percent accuracy across 392 secure environments. Ashish Kumar, Coforge’s global head of cloud and AI infrastructure, frames the stakes plainly. What limits autonomy is not intelligence but the operational foundation beneath it.
Strip the announcement down, though, and a more interesting story emerges. Coforge, a $1.4 billion IT services firm, is selling agentic AI operations — the discipline of keeping agents safe, accurate, and affordable — as billable work. The problems it cites are real. The packaging is worth a closer look.
The Competitive Picture for Agentic AI Operations
Three forces shape this launch: the documented failure modes of production agents, a fast-growing tooling market attacking the same problems with software, and the business pressures on services firms themselves.
The problem Coforge sells against is real
Gartner expects more than 40 percent of agentic AI projects to be canceled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls — the exact gaps Coforge lists. The analyst also estimates only about 130 of the thousands of self-described agentic vendors are real, calling the rest “agent washing.”
The cost pain is quantified. Production measurements show multi-agent systems consuming roughly 15 times the tokens of equivalent single-agent interactions. Tool schemas alone can account for 60 to 80 percent of token usage in static toolsets. Organizations running five or more agents report 35 to 50 percent higher costs than projected. One documented agent retry loop burned $187 in ten minutes.
The tooling race is already funded
Agentic AI operations is not just a services opportunity — it is a software category. The LLM observability market reached roughly $2.7 billion in 2026, up 36 percent from the prior year. LangSmith, Braintrust, Arize Phoenix, Langfuse, Helicone, and AgentOps all sell tracing, evaluation, drift detection, and cost attribution. Datadog and New Relic now treat LLM steps as traceable spans.
So buyers of agentic AI operations have a build-or-buy choice Coforge’s release never acknowledges. Why rent a services-led harness when a product-led stack, instrumented on OpenTelemetry conventions, plugs into tooling the enterprise may already own?
The services-firm stakes
Coforge needs this narrative. Its FY25 revenue reached $1.45 billion, up 31.5 percent — but a chunk of that came from acquiring Cigniti, and IT services growth overall has been slowing industry-wide. CEO Sudhir Singh has said the firm is “within touching distance” of $2 billion. Agentic AI threatens the labor-arbitrage model that funds every services firm. Selling the operations of the agents that would otherwise replace billable people is a credible hedge.
What the Data Shows
Coforge’s financial position supports the bet. FY25 closed with a record $3.5 billion order intake, including a $1.56 billion total-contract-value deal signed in the fourth quarter. The executable order book stands at $1.5 billion, up 47.7 percent year over year. Headcount sits near 33,500, with attrition at 10.9 percent — among the industry’s lowest.
Against that backdrop, the AgenticOps claims deserve scrutiny. The release cites 98.2 percent accuracy for 37 clients across 392 secure environments. Accuracy of what, measured how, against what baseline?
The release does not say. No client is named. No methodology appears. The phrase “significantly faster threat remediation” carries no number at all.
The claim pattern repeats across Coforge’s own announcements. Earlier EvolveOps.AI releases cited 25 percent downtime reduction, 40 percent IT-cost savings, and 60 percent faster detection and resolution. Those too came without named customers or methodology.
What’s New vs. What’s Repackaged
Genuinely new
The framing is fresh: positioning agent governance, drift correction, and token optimization as a continuous operations discipline rather than a deployment feature. The emphasis on token economics lands well — cost is where agent programs die.
Improved
EvolveOps.AI itself keeps gaining scope. Each announcement adds capabilities — most recently the governed registry and the explicit drift-detection story.
Repackaged
The platform has now had four announcements in nine months, each re-positioning the same core: launched December 2025 as an agentic IT-operations platform, “advanced” in February 2026 with identical metrics, folded into the Nuuron AI operating system in July, and now recast as an AgenticOps layer. AIOps became an AI-OS became AgenticOps. The names chase the market’s vocabulary faster than the product changes.
Unclear
What the headline metric measures. Who the 37 clients are. What the harness costs, or whether it is sold as a product, a subscription, or billable hours. And the release expands MCP as “model control protocols” — the standard is the Model Context Protocol, a naming slip a security-and-governance pitch should not make.
The Question That Wasn’t Answered
Is this a product or a services wrapper? A harness you can buy, evaluate, and cancel — or a branded methodology that routes to Coforge consultants? The release never separates the two, and the answer determines everything about pricing, lock-in, and how the 98.2 percent claim could ever be audited.
A second silence: the cannibalization question. If AgenticOps succeeds, clients eventually run leaner with fewer billable people. A services firm selling autonomy must eventually sell against its own delivery model. Coforge’s FDE-and-pods structure suggests an answer — but the release does not make it. Agentic AI operations sold as a black box is not operations; it is faith.
What Agentic AI Operations Buyers Should Ask
If you run an AI program, take the problem seriously and the pitch provisionally. Demand the methodology behind any accuracy figure, and a reference client who will speak to it.
If you are choosing a stack, price the product route first. LangSmith, Arize Phoenix, Langfuse, and Datadog already sell tracing, evals, and cost attribution on OpenTelemetry conventions — some self-hostable for data-sovereignty needs. A services harness must beat that baseline.
If you are an IT services competitor, note the pattern. Coforge is converting the industry’s existential threat into its next practice line. Expect every large peer to announce the equivalent within a year, and expect the differentiation battle to be fought on audited outcomes, not announced percentages.
Editor’s Note
This article draws on Coforge‘s September 2026 press release, its prior EvolveOps.AI and Nuuron announcements, FY25 regulatory results, and public analyst and industry research on agentic AI. Performance figures cited for EvolveOps.AI and AgenticOps are company-reported and independently unaudited. No product was evaluated hands-on.