MyJackfruit

Help
Browse topics · Agents and AI

Agents and AI

← Agents and AI

My AI does not do what I ask. What now?

Short answer

Check the right agent is chosen, say it again in plainer words with an example, and name your project's vault so it reads your project.

Before you start

Words used here (2)
agent
An agent is your AI following one written procedure, its standing instructions, for one kind of work.
project
A project is work to bring about a change you want, such as a new app or a better way of doing something.
Deeper

Also asked:

My AI does not do what I ask during setup. First check that your AI can reach your vault, and that Claude Code or Grok shows with Disconnect under it in GilbApp's chat. Then say it again in plainer words, with an example.

The agent answered but did nothing. What now? Read its answer: Kai-Zen may be waiting for your reply to its questions, or your yes to its plan. If not, check that your AI can reach your vault, then say plainly what to do and in which vault; files on your computer need a code folder or Claude Code in Terminal.

While you set up, first check that your AI can reach your vault; the answer on checking that your AI can reach your vault shows how. In GilbApp's chat, to the right of the agent list, your AI's name, Claude Code or Grok, shows with Disconnect under it; Connect there means none is connected, and connecting has its own answer. Without your own AI, the chat answers only if GilbPlatform has switched on its shared AI. GilbApp's chat does not reach the files on your computer unless you choose a code folder while Claude Code is connected; otherwise anything with files or a folder there needs Claude Code in Terminal, the window where you type commands (on a Mac, the Terminal app). The answer How do I let my AI build in a folder on my computer from GilbApp's chat? shows how.

Check which agent you are talking to. An agent is your AI following one written procedure, its standing instructions, for one kind of work. Kai-Zen plans, Jackfruit works in your vaults, Agent.Concept names kinds of data, and Agent.Coder builds apps. If you are not sure which agent to ask, ask Kai-Zen. The answer on which agents there are tells more. If a line above the box says The agent is answering WITHOUT its standing instructions, or with only part of its standing instructions, the agent is answering without all of its written procedure, and its answer may not follow its way of working. Try again later; if the line stays, write to [email protected].

When the AI misunderstands you, tell it so in the same chat: what you meant, an example, and what it got wrong. Short, plain sentences help. Kai-Zen asks what only you know in one message at the start; what else it had to choose, it writes down as an assumption, a choice it made for you that you can change. At the end it shows you the whole result, with every assumption, for one approval. Reply with what you want changed; it changes it and asks once more. A change you want later goes as the answer on changing the product while a project runs shows.

When the AI has lost track of your project: GilbApp keeps a chat in its browser tab only, and only its last 32 messages and replies, so earlier messages drop out. What lasts is what the agent saved where the project is kept. Kai-Zen does not need your earlier chat: name the project's vault and say what you want. Before it does anything, it reads the project's setup, its specs and its last round of delivery, as the answer on continuing a project with Kai-Zen shows. It does not ask again what you already answered: it looks first in what the project has written down. End each session by typing close, so Kai-Zen writes a handoff, the note on what was done, what is open and where to start next time; the answer on ending a session shows how.

When it keeps making the same mistake. Say plainly what it keeps doing, what you want instead, and give one example of the right result. If it is a fault in what was built, the answer How do I tell my AI about a fault so it can fix it? shows what a good report says. Agent.Coder stops by itself after three different failures it cannot fix without more scope or your input, and tells you what failed and what it tried: answer it with what you know. If the same mistake comes back in each new session: Claude Code itself starts each session with a fresh context window, what it can keep in mind at once; Anthropic says two things carry knowledge across sessions: CLAUDE.md files, which you write, and auto memory, notes Claude writes itself (https://code.claude.com/docs/en/memory, read 13 November 2026). If it changed things you did not ask for, see My AI changed things I did not ask for. What now?

Anthropic says Claude Code in Terminal saves every conversation on your computer, and that this line starts the last one in the same folder again (https://code.claude.com/docs/en/common-workflows, read 13 November 2026):
claude --continue

If nothing helps, write to [email protected]: what you asked, what the AI did, and at what time.

This answer is written for a computer screen.

Source: GilbApp 96968de7c3e3e685296f20a2599cbc8d305c9ea7 (live build, read 12 November 2026): src/components/agent/AgentChat.tsx:67, :129 (a chat keeps its last 32 turns, in the tab's session storage); src/lib/base/agentChat.ts:128-142 (the standing-instructions lines). GilbPlatform 670005ccd64c229c59f905a00430d906781eb942: src/routes/ai-chat.ts:231 (shared AI refusal). Standards Kai-Zen.AskRules (One, Assume, Settled, EndApproval), Kai-Zen.Existing (as cited by the live page on continuing) and Kai-Zen.Close, read from GilbPlatform on 12 November 2026. Claude Code: https://code.claude.com/docs/en/common-workflows Resume previous conversations (read 13 November 2026). Shared wording: live answers How do I check that my AI can reach my vault?, How do I start work with Kai-Zen?, Which agents are there, and which do I use for what?, Why does Kai-Zen stop and ask, or follow a standard?, How do I answer Kai-Zen's questions?, How do I continue a project with Kai-Zen?, How do I read what an agent hands back?, How do I end a session?, Where do I get more help?. Agent.Coder (AfterIncrement: Stop after three distinct failures ... Report each attempt). Anthropic: https://code.claude.com/docs/en/memory. Live answers, sentences shared: How do I tell my AI about a fault so it can fix it?; What can my AI see and do in my vault, and does it remember? Code-folder exception: GilbApp 49b1ae53 (https://gilb.com/app/version.json, read 13 November 2026), src/components/agent/AgentChat.tsx:677-692 and :1084-1090 (Choose a code folder, Code in, Claude Code only); live answer How do I let my AI build in a folder on my computer from GilbApp's chat?

Checked against GilbApp d529620.