AI agents are computer programs that can plan and act to reach a goal with little or no human help. Gartner says AI agents could change work in many jobs. At the same time, many leaders feel let down by early results. This mix of promise and doubt is normal for a new tool. It helps to know what AI agents can do today, what still blocks progress, and how to invest wisely.
How AI agents could change work
AI agents aim to move beyond simple chat. A user gives a clear goal, and the agent breaks it into steps, calls the right tools, and checks results. Gartner’s analysts explained at its 2024 conference that today’s chat systems are passive, while agentic AI will work from high‑level goals and run action steps on its own. Early business uses include support tickets, marketing content, code help, IT ops, and security triage. Over time, firms want agents to run complex, multi‑step processes across teams.
What “agentic” means, in plain words
Agentic AI means the system can decide next actions toward a goal. It keeps short‑term memory, plans, and uses tools such as search, email, or databases. It still needs guardrails and review.
Gartner hype cycle: why disillusionment is normal
Gartner maps new tech on a “hype cycle.” Interest spikes, then dips into the “trough of disillusionment” when results lag. Multiple reports in 2024 showed gen AI moving into this dip. Analysts said leaders now ask harder questions about value, safety, and change effort, not just demos. One summary reported that generative AI is sliding into the trough of disillusionment. Another noted the same shift as Gartner shared its 2024 cycles for AI and emerging tech, with buyers pushing for real ROI and fewer promises (coverage of the shift).
What this means for AI agents
Because agents build on gen AI, they ride the same curve. Some vendors now “agent‑wash,” renaming simple bots as agents. That raises hopes, then disappoints. Teams should test small, honest use cases, publish the value metrics, and avoid flashy claims.
Enterprise AI costs: the main roadblock
Cost is a common pain. Gartner told CIOs that model use, data prep, and new app licenses can add up fast, and many firms underestimate spend. Reports based on Gartner surveys said that more than 90% of CIOs see cost control as a barrier to AI value. Leaders also warn that some vendors raised prices as they bundle AI into existing tools. To avoid waste, teams should set clear cost limits, track per‑task cost, and push workloads to right‑sized models.
Practical cost moves
- Start with narrow jobs where errors are cheap and wins are easy to count.
- Use small or open models for routine work; reserve large models for hard tasks.
- Cache, batch, and reuse results to cut token spend.
- Treat AI like a portfolio: retire weak pilots; scale only when value is proven.
Related reading: our review of Humane’s Ai Pin hype and outcome shows how early excitement can outrun real‑world results.
Agentic AI maturity: what to expect through 2028
Gartner’s later guidance in 2025 warned that many “agentic AI” projects will be dropped for weak value and high complexity, and that vendors sometimes mislabel tools. A June 2025 report noted that over 40% of agentic AI projects may be scrapped by 2027. At the same time, Gartner expects growth: by the late 2020s, a larger share of apps may include agent features, and some routine decisions may be automated. The take‑home: progress is real, but steady, not magic.
Where agents work best today
- Internal help: search your company docs, draft answers, file tickets
- Marketing ops: repurpose content, check tone, schedule posts
- Dev and IT: code suggestions, log triage, test generation
- Security: alert summaries, playbook steps with human review
See also: examples of AI used for complex communication, such as marine‑mammal call work that used machine learning, show the power and limits of today’s systems.
How to pilot AI agents safely in 2025
- Pick one high‑value job and write a clear success score (time saved, tickets closed, errors caught).
- Use a “human in the loop” for checks until the error rate is near zero.
- Log every action the agent takes; keep traceable steps.
- Add memory and tools only when the agent proves safe with fewer.
- Plan rollback: if cost or quality drifts, the agent stops and hands back to a person.
Limitations & quality of evidence
This article summarizes analyst talks and media reports, not a single peer‑reviewed study. Gartner’s numbers are forecasts, not measured facts. Real adoption will vary by task, data quality, and change management.
VentureBeat – Gartner predicts AI agents will transform work – 2024
At Gartner’s 2024 Symposium/Xpo, analysts said today’s chat systems are passive, while AI agents will take high‑level goals and run action steps. The piece also notes rising costs and early signs of disillusionment.
Computerworld – Generative AI is sliding into the trough of disillusionment – 2024
A report on Gartner’s 2024 hype cycles explained that generative AI is moving into the trough of disillusionment, as buyers seek clearer value and safer use.
Reuters – Over 40% of agentic AI projects will be scrapped by 2027 – 2025
Reuters covered a 2025 Gartner note saying that over 40% of agentic AI projects may be canceled by 2027 due to cost and unclear value. It also reported Gartner’s view that agentic AI will still grow.
Gartner – The 2025 Hype Cycle for GenAI Highlights Critical Innovations – 2025
Gartner’s 2025 overview explains that organizations want to use AI agents to automate complex, multi‑step processes at scale. This sets realistic aims for near‑term agent projects.
0 Comments