Effect AI patterns
Effect.gen(function* (topic: string) {
const ideas = yield* brainstorm(topic);
const scores = yield* critique(ideas.join("\n"), ["novelty", "feasibility"]);
const topPick = yield* consensus(scores);
const plan = yield* plan(`Implement idea ${topPick.index}`);
})
Living idea list:
recall(context): a pattern for adding context to the current conversation without completely breaking the flow of the conversation. Should potentially use an LM in a strand to decide the best way to phrase the additional context and then reemphasize the focus of the conversation.explore_alternatives(prompt): From a given prompt or situation, the conversation spawns alternate dialogues – each branch might assume a different approach, parameter, or interpretation – and generates outcomes for each.goal_oriented(target): change the tone of / update the target (maybe some text, maybe a runic?) such that all LM replies are geared towards a clear goal or desired outcome.classify(input): Categorizes / labels the input. Perhaps there are common classifications that make sense (topic, sentiment, urgency, etc.). Perhaps it should be left configurable or open to LM interpretation.brainstorm(topic): generate distinct ideas based on a given topic. Kind of similar to expand. Might be unnecessary.summarize(source): Produces a concise summary of a longer text, transcript, or multi-turn history.critique(content, criteria[]): Evaluates content against an explicit checklist.consensus(child_results[]): Aggregates multiple answers (from debate() or explore_alternatives()) into a single, best-fit resolution.plan(goal): Breaks a high-level goal into ordered steps with owners and success metrics.simulate(role_play_spec): Runs a multi-agent role-play to test scenarios, dialogs, or negotiations.generate_files(tree_desc)edit(file_path, edit_desc)
Effect AI patterns
Living idea list:
recall(context): a pattern for adding context to the current conversation without completely breaking the flow of the conversation. Should potentially use an LM in a strand to decide the best way to phrase the additional context and then reemphasize the focus of the conversation.explore_alternatives(prompt): From a given prompt or situation, the conversation spawns alternate dialogues – each branch might assume a different approach, parameter, or interpretation – and generates outcomes for each.goal_oriented(target): change the tone of / update thetarget(maybe some text, maybe a runic?) such that all LM replies are geared towards a clear goal or desired outcome.classify(input): Categorizes / labels the input. Perhaps there are common classifications that make sense (topic, sentiment, urgency, etc.). Perhaps it should be left configurable or open to LM interpretation.brainstorm(topic): generate distinct ideas based on a given topic. Kind of similar toexpand. Might be unnecessary.summarize(source): Produces a concise summary of a longer text, transcript, or multi-turn history.critique(content, criteria[]): Evaluates content against an explicit checklist.consensus(child_results[]): Aggregates multiple answers (from debate() or explore_alternatives()) into a single, best-fit resolution.plan(goal): Breaks a high-level goal into ordered steps with owners and success metrics.simulate(role_play_spec): Runs a multi-agent role-play to test scenarios, dialogs, or negotiations.generate_files(tree_desc)edit(file_path, edit_desc)