Agentic AI: Hype or revolutionary?
In general, I stay away from people who talk too much about "Agentic AI". It's such a hyped word, and I know what's behind it - automation with an LLM API call. And the goal should never be "use agentic AI". That said, agentic AIs can be helpful - only when it's the best solution for the problem you are trying to solve.
When deciding whether to adopt AI tools or “agents,” you shouldn’t start with the tools themselves. You should start by examining the problem. Not every workflow benefits from AI, and forcing it often adds complexity instead of reducing it. To help evaluate whether Agentic AI is a good fit, here is a practical checklist you can use for your own business or team.
What makes a good candidate for AI or automation
- Repetitive, high-volume, or time-consuming tasks — data entry, email triage, report generation, summarization, or simple customer inquiries. These tasks don’t require complex logic, and automation can free significant time.
- Rule-based or pattern-based work (with tolerance for small errors) — formatting, moving files, consolidating documents, summarizing logs. AI’s ability to interpret natural language can outperform classic RPA in flexible situations.
- Clear metrics for success — time saved, errors reduced, speed improved, or cost lowered. If outcomes are hard to quantify, the investment may not be justified.
- Low complexity / minimal branching logic — tasks with few exceptions work better. Multiple edge cases and deep logic make agentic AI behave unpredictably.
- Non-core, non-critical tasks — internal operations, admin work, data consolidation, drafting, and back-office support benefit the most.
- You have someone who can verify the output — AI results are probabilistic, not deterministic. Think of AI as a highly capable intern: helpful, but not ready to operate completely unsupervised.
When you should avoid AI or Agentic automation
- The task requires perfect consistency, compliance, or accuracy — financial calculations, regulatory workflows, legal documentation. Most of the time, traditional engineering solves them better.
- Your goal is unclear or inconsistent — automation is useful only when you know the desired outcome.
- Success is hard to measure, or errors are costly — keep the task human-driven, or go for traditional engineering.
- Human judgment or empathy is essential — domain knowledge, relationship management, sensitivity. AI can support but not replace these responsibilities.
- The task is rare or one-off — building automation for a task that rarely happens usually is not worth the time.
Why this matters
Before adopting agentic AI (or any AI), define the problems really well. The right approach is: define the real problem to solve, understand the desired outcome, then decide whether AI adds meaningful value.
Don't go with an "I have to use AI" mindset - I've heard many of the non-tech people say this, but it will not lead to any positive outcome. Also, there are plenty of IT service providers who use "agentic AI" in their pitch - focus on the discussions around problem solving, and not solutions. If you meet a vendor who can honestly say whether AI makes sense in solving your pain or not, it can be a good criterion for deciding whether they are the right partner.
You don’t need AI for everything. You need AI where it saves time, reduces cost, or frees people to focus on meaningful work. Sometimes, the solution is not AI. Use AI only when it matters.