diff --git a/src/pages/autocody/index.tsx b/src/pages/autocody/index.tsx
index 2819a8948d3..56c781e6209 100644
--- a/src/pages/autocody/index.tsx
+++ b/src/pages/autocody/index.tsx
@@ -25,7 +25,7 @@ const AutoCodyPage: FunctionComponent = () => (
- We are building our autonomous coding assistant that will be able to write code - for you, understand your problems, and guide you in solving them. + We are building an autonomous coding assistant that will be able to understand + your problems, create a step-by-step plan to solve them, and write code + alongside you.
+ +Principle 1: Hybrid
+Principle 1: Cooperation
- A combination of coder and AI. There are agentic - solutions that let the AI free to choose the correct path to solve an issue from start to finish.{' '} + A combination of human coder and AI. Agentic solutions, + like{' '} SWE-Agent - , the benchmark for autonomous code generation, is only able to achieve a 12.29% success rate on the - full SWE-bench test set. If you want to create an autonomous car, you don't remove the steering - wheel. We must allow the AI to be trained and "steered" by a user. + , that allow the AI to define a plan and execute against it fully autonomously are only able to + achieve a 12.29% success rate on the full SWE-bench test set. If you want to create an autonomous + car, you don't remove the steering wheel. We must allow the AI to be trained and "steered" by a + user. +
++ We believe the best place for a user to provide high fidelity feedback to AI is in a code editor.
@@ -281,8 +302,8 @@ function Principles(): JSX.Element {
Principle 2: Context
- Context should come from inside AND outside a repo. We - are building{' '} + Context should come from inside AND outside the codebase. {' '} + We are building{' '} OpenCtx - , an open source tool to bring in content about your code from sources like Linear, Jira, Slack, - Google Docs, and more. We believe that an LLM (and a human) can't effectively solve a problem - without the context of the problem. + , an open source tool to bring relevant context about your code from sources like Linear, Jira, + Slack, Google Docs, and more. We believe that an LLM (just like a human) cannot effectively solve a + problem without full context and understanding of the problem.
Principle 3: Steerable
A user should be in the loop at every step. In an agentic - workflow, AI will have many steps to complete for the user. A user should be able to steer the - workflow at every step by evaluating LLM output throughout the process. + workflow, AI will do the work, with the human providing oversight. A user should be able to steer + the workflow at every step by evaluating LLM output throughout the process.
Principle 4: Context
-Principle 4: Feedback
++ AI should learn from feedback. AI should continuously + evolve and adapt to better understand the user's intent, not just the code. Gathering feedback at + every step of the process. +