Trying Again
Have you ever tried to open a jar with a tight lid? You twist and twist, but it will not budge. So you try a different way — maybe you run it under warm water, or ask someone stronger for help, or use a rubber jar opener. You did not give up. You tried again, differently. AI agents do exactly the same thing. When an action does not work, a good agent does not just freeze or quit. It checks what happened, thinks of a different approach, and tries again.
Why the Same Action Might Not Work Twice
The world is complicated. The same action that worked yesterday might not work today because something changed. Imagine a robot delivering mail through a building. It rolls down the hallway toward Room 3. But today someone left a large cart blocking the hallway. The robot cannot go straight anymore. A robot that just repeats the same action — roll forward, roll forward, roll forward — will crash into the cart over and over. That is not useful. A smart agent checks what happened (sensed the cart blocking the path), thinks of a different way (maybe go around the cart by taking a different hallway), and acts differently. It tries again — but with a new plan.
When an action does not work, a good agent does not repeat the exact same action and hope for a different result. It senses what went wrong, thinks of a new approach, and tries again with a better plan.
Here is a story about trying again in the digital world. An agent is looking for a piece of music online. It searches for the exact title and gets zero results. What does it do? It does not search the same words again — that would give the same zero results! Instead, it thinks: maybe the title is slightly different. It tries a search with alternate words — maybe just the artist's name. It gets results! It finds the music. The agent tried again with a smarter approach. That is intelligent trying-again, not just stubborn repeating.
Match each failed action to the smarter retry the agent should try.
Terms
Definitions
Drag terms onto their definitions, or click a term then click a definition to match.
How does an agent know to try something different? This is where thinking matters most. After sensing that an action failed, the agent thinks about WHY it failed. Was the path blocked? Was the information wrong? Was the grip too tight? Each different reason leads to a different kind of retry. A truly smart agent does not just try random things. It uses what it sensed to make a better guess about what to do next. Every failed action is information that helps the agent choose a smarter retry.
When an agent's action fails, that failure is not wasted. It is valuable information! The agent now knows one thing that does not work, which helps it figure out what might work instead. Every failure makes the next try smarter.
Flashcards — click each card to reveal the answer
A robot tries to roll through a door but the door is closed. What should the agent do next?
Why does an agent try DIFFERENTLY the second time instead of the exact same way?
The Retry Challenge
- Play this problem-solving game with one or two friends.
- One person is the WORLD. They pick a goal — for example: get a small toy from across the room without walking directly to it.
- The WORLD makes a rule: one path is blocked (you cannot go straight). They do not tell the agent which path is blocked — the agent has to find out by trying.
- The other person is the AGENT. The agent tries an action. The World says: blocked! or works!
- If blocked, the agent must try a different route.
- Keep going until the agent reaches the goal.
- After the game, the agent explains: what did they learn from each blocked attempt that helped them find the working path?
- Swap roles and play again with a new goal and new blocked paths.