# Architect Agent Uses Grill-Me to Ask Better Questions > Source: > Published: 2026-06-03 13:01:46+00:00 # Architect Agent Uses Grill-Me to Ask Better Questions ### Instead of turning vague prompts into confident plans that won’t work, the Architect agent now stops to interrogate the task before it writes the plan. Most planner modes fail in a familiar way: they start and end planning too early. They take an underspecified request, fill in the blanks themselves, and produce a plan that looks polished but rests on bad assumptions. Existing plan modes on the market usually deliver plans that aren’t well defined and lack refinement. Another behavior users weren’t happy about is how eager the agent is to implement the plan - and the continuous push to implement can be an anxiety trigger for some. That’s the problem we wanted to fix. ## What changed We’re introducing a new Architect agent in the [Kilo](https://kilo.ai/) Marketplace, whose core was inspired by the Grill-Me skill. It leverages instructions that make the agent challenge the user instead of handing off a half-baked plan after just a couple of questions and committing to an implementation. Instead of pretending the request is clear, when there’s room to disambiguate, it asks the questions a good engineer would ask first: what matters, what’s constrained, what can be deferred, and what success actually looks like. The goal is to limit conversation, and deliver a plan that is well defined, so you can actually execute. It’s worth noting that the number of questions and their quality depend heavily on how you define the first prompt and how you answer each one. In my experience, almost intuitively, it keeps asking questions as long as I want it to, based on the inputs I give. For instance, I can steer the direction by providing custom answers and adding or removing scope. ## What is Grill-Me? [Grill-Me](https://github.com/mattpocock/skills/blob/main/skills/productivity/grill-me/SKILL.md) is a skill created by Matt Pocock, and it became very popular because it really produces incredible results. What’s interesting about this skill is that it’s so concise - about four lines of text - while still being very effective. The idea is that it will challenge your initial ideas from several angles until they become a more refined output. ## A concrete example Take a prompt like: build planner mode for Architect. That sounds clear until you start pulling on it. *Should it ask questions every time, or only when ambiguity is high?* *Should it prefer repo inspection before asking?* *Should it read the repo documentation up front, or wait for the user to ask for it explicitly?* *What counts as enough clarity to move from questioning to planning?* *What does a good output plan actually need to contain?* Grill-Me forces those decisions into the open, and the resulting plan is usually narrower, more concrete, and much easier to implement. ## Why it matters A lot of bad implementation work is just unresolved ambiguity with code wrapped around it. If planner mode can surface the missing decisions earlier, the rest of the workflow gets better: fewer wrong assumptions, less rework, and a better handoff to Code mode. The value of Architect planner mode is not that it writes a long plan. It’s that it knows when not to write one yet. ## How to use Architect ### Head to the Kilo Marketplace. ### Install the Architect agent. ### Select Architect in the agent mode picker. ### Give your agent an objective, and watch it get clarity on your plan. ### Decide if it’s time to finalize the plan or keep refining. ## Getting started with Kilo Code #### 500+ models. One open source agent in [VS Code](https://kilo.ai/install), [JetBrains](https://kilo.ai/features/jetbrains-native), [CLI](https://kilo.ai/cli), [Cloud](https://kilo.ai/cloud). Kilo Code is available across your favorite environments. Here’s how to get up and running in minutes: : Search for Kilo Code in the VS Code Extensions Marketplace, install it, and sign in. You’ll find Kilo in the sidebar, ready to go.[VS Code](https://kilo.ai/install): Open Settings → Plugins, search for Kilo Code, install it, and restart your IDE. Kilo will appear as a tool window on the side.[JetBrains](https://plugins.jetbrains.com/plugin/28350-kilo-code): Run npm install -g kilocode (see[CLI](https://kilo.ai/cli)[all install options](https://kilo.ai/cli)), then authenticate with kilocode login. Use kilocode from any terminal to start coding with AI. Once you’re in, select your [model](https://kilo.ai/leaderboard) and start building. ## Tell us what you think I would love to know how it works for you. Try the Architect agent on a real task and tell us whether the questions surface the right decisions, whether it knows when to stop and plan, and where it gets in the way. We have plans to iterate further on this, so your feedback directly shapes what we ship next. Reach out at [hi@kilocode.ai](mailto:hi@kilocode.ai) or on our Discord at [kilo.ai/discord](http://kilo.ai/discord).