Agentic AI is AI that doesn't just answer one message and stop — it takes a series of steps on its own toward a goal you gave it, deciding along the way what to do next. Instead of "ask, get one answer," it's closer to "ask, and it keeps working until the job is actually done."
A regular chatbot responds once per message and waits for you. Agentic AI can plan a sequence of actions, check its own progress, and keep going without you re-prompting it at every step. The difference is passive Q&A versus something that actually pursues an outcome over time.
Because the underlying models got reliable enough to string multiple steps together without falling apart, and that unlocks a much bigger category of usefulness — going from "AI that talks" to "AI that gets things done." It's the current frontier the whole industry is racing toward, from big AI labs to everyday consumer apps.
Instead of asking an assistant "what's a good gift for my sister" and stopping there, agentic AI could be told "find and suggest three gift ideas for my sister's birthday next month, and remind me to order one a week before" — then it plans, tracks the date, and follows through without you having to babysit every step.
They're related concepts but different products. Developer-built AI agents are typically custom code, APIs, and frameworks that engineers wire together for a specific automated workflow. Agentic AI in a consumer app like HAL is that same core idea — multi-step, goal-directed AI — but packaged into a chat app you use directly, no code or API key required.
It used to be mostly a developer thing, but that's changing fast. HAL is a real example of agentic AI built for regular people: you set up a Routine or a Goal in plain language in the chat app, and HAL handles the multi-step part in the background — nothing to code, configure, or deploy.
Routines are recurring automated tasks — tell HAL to do something on a schedule, like a daily news roundup or a weekly summary, and it keeps doing it without you asking again. Goals are longer, multi-step pieces of work — you give HAL an outcome you want, it keeps working on it in the background over time, and delivers the result to your inbox when it's done. Together they're HAL's real agentic capability.
In HAL, you're always the one who sets the goal or routine in the first place — nothing runs unless you tell it to, and you can see and manage what's active at any time. Agentic AI acting "on its own" means working through the steps of a task you assigned, not making independent decisions about your life or accounts.
A Routine repeats — it's something you want done again and again on a schedule, like "every Monday, summarize what happened in AI news." A Goal is a one-time, longer piece of work with an end point — like "research and put together a comparison of three options and send it to me when it's ready." Routine is about repetition; Goal is about depth on a single outcome.
It's not a fully autonomous system that can act anywhere on the internet or handle every real-world task unsupervised — it's still bounded by what the tools around it can actually do, and it works best on well-defined, specific goals rather than vague, open-ended ones. Consumer agentic AI like HAL's Routines and Goals is genuinely useful today, but it's a capable assistant carrying out tasks you define, not a general autonomous agent doing whatever it decides on its own.