Agentic AI
Agentic AI projects: how to pick a first one and keep it on track
A plain guide for owners and IT leads, based on Gartner, Anthropic and Microsoft, read 11 November 2026.
Agentic AI projects are efforts to build AI systems that carry out multi-step work with some freedom to decide how. A Gartner press release dated 25 June 2025, read on 11 November 2026, predicts that over 40% of agentic AI projects will be canceled by the end of 2027, due to escalating costs, unclear business value or inadequate risk controls. That is a prediction, not a measurement of any one team. The three causes are things a team can address before starting: write down the value, estimate the running cost, and decide the risk controls. Then choose the simplest design that works.
plain answer
What are agentic AI projects?
Agentic AI projects are projects that build systems in which an AI model directs its own steps and tool use. Anthropic's article "Building effective agents", read on 11 November 2026, defines agents as systems where models dynamically direct their own processes and tool use, and contrasts them with workflows, where models and tools are orchestrated through predefined code paths.
A project that carries the label may turn out to be a workflow, not an agent. That is fine. Anthropic recommends finding the simplest solution possible and increasing complexity only when needed, which might mean not building an agentic system at all.
how to
Why do agentic AI projects get canceled?
For agentic AI projects, Gartner's June 2025 press release gives three reasons: escalating costs, unclear business value and inadequate risk controls. Its headline prediction is that over 40% of agentic AI projects will be canceled by the end of 2027. We quote the prediction as Gartner states it. It is a forecast, and we did not read Gartner's underlying research.
Anthropic adds a cost point from the design side: agentic systems often trade latency and cost for better task performance, so that trade should be worth it for the task.
how to
How do you pick a first agentic AI project?
These steps are Kriv's suggestion, built from the three reasons above, and are not a quotation from any source.
Value: name the one result the project should change, and who in the business will judge it. If nobody can say, do not start yet.
Cost: estimate what a run costs and how many runs you expect, using your own model and tool prices. Do not borrow a number from a vendor page.
Risk controls: list what the system may read, what it may change, and which actions need a person to approve. The agent security risk guide on this site covers the common risks.
Size: choose a task that is narrow and has a clear end, so you can tell whether it worked.
how to
Which design should an agentic AI project use?
Use the least complex design that reliably meets the requirement. Microsoft's Azure Architecture Center, read on 11 November 2026, describes three levels: a direct model call, a single agent with tools, and multiagent orchestration. It calls a single agent with tools often the right default for enterprise use cases and says multiagent setups add coordination overhead, latency and failure modes.
Anthropic says optimizing single model calls with retrieval and in-context examples is usually enough for many applications. If you do use a framework, it advises making sure you understand the underlying code.
Our multi agent workflow guide explains when several agents are worth the added complexity.
kriv role
Where Kriv helps
Kriv AI helps teams scope, set up and govern agentic AI work, including the value, cost and risk questions above. See agentic AI governance, agentic AI architecture and agentic AI training for how that work is described on this site. Kriv does not speak for Gartner, Anthropic or Microsoft, and none of them has reviewed this page.
Your own leaders hold the final say on what is built.
limits
What this page could not confirm
We read the Gartner press release page, Anthropic's article and Microsoft's guide once each on 11 November 2026. We did not read Gartner's research behind the prediction, and we cite no other figure. We list no example projects, because we did not find a source we could quote for them. We make no claim about results for your project, and nothing here is legal advice.
Related resources
Continue exploring
Straight answers
Frequently asked questions about Agentic AI projects: how to pick a first one and keep it on track
What is an agentic AI project?
A project that builds a system in which an AI model directs its own steps and tool use, as Anthropic defines agents in its article read on 11 November 2026.
Will most agentic AI projects fail?
Gartner predicts that over 40% will be canceled by the end of 2027, due to escalating costs, unclear business value or inadequate risk controls. That is a forecast, not a count of finished projects.
Should our first project use an agent at all?
Maybe not. Anthropic says the simplest solution possible is best and that this might mean not building an agentic system. Microsoft calls a single agent with tools often the right default.
What should we decide before we start?
The value you want, a cost estimate from your own prices, and the risk controls, including where a person approves. This is Kriv's suggestion and not a vendor statement.
Can Kriv help us scope one?
Kriv AI helps teams scope, set up and govern agentic AI work. Your own leaders decide what is built.
Talk to the team that would do the work
Bring your requirements to a working session with the person who'll actually deliver.
Book a Discovery Call