diff --git a/packages/ai/src/agents/think-and-prep-agent/get-think-and-prep-agent-system-prompt.test.ts b/packages/ai/src/agents/think-and-prep-agent/get-think-and-prep-agent-system-prompt.test.ts
new file mode 100644
index 000000000..ae5de9b9a
--- /dev/null
+++ b/packages/ai/src/agents/think-and-prep-agent/get-think-and-prep-agent-system-prompt.test.ts
@@ -0,0 +1,30 @@
+import { describe, it, expect } from 'vitest';
+import { getThinkAndPrepAgentSystemPrompt } from './get-think-and-prep-agent-system-prompt';
+
+describe('getThinkAndPrepAgentSystemPrompt', () => {
+  it('should return system prompt with SQL dialect guidance', () => {
+    const sqlDialectGuidance = 'PostgreSQL specific guidance';
+    const result = getThinkAndPrepAgentSystemPrompt(sqlDialectGuidance);
+
+    expect(result).toContain('You are Buster, a specialized AI agent');
+    expect(result).toContain('PostgreSQL specific guidance');
+    expect(result).toContain("Today's date is");
+  });
+
+  it('should include all necessary sections', () => {
+    const sqlDialectGuidance = 'MySQL specific guidance';
+    const result = getThinkAndPrepAgentSystemPrompt(sqlDialectGuidance);
+
+    // Check for key sections
+    expect(result).toContain('<intro>');
+    expect(result).toContain('<prep_mode_capability>');
+    expect(result).toContain('<event_stream>');
+    expect(result).toContain('<agent_loop>');
+    expect(result).toContain('<todo_list>');
+    expect(result).toContain('<todo_rules>');
+    expect(result).toContain('<tool_use_rules>');
+    expect(result).toContain('<sequential_thinking_rules>');
+    expect(result).toContain('<execute_sql_rules>');
+    expect(result).toContain('<sql_best_practices>');
+  });
+});
\ No newline at end of file
diff --git a/packages/ai/src/agents/think-and-prep-agent/think-and-prep-instructions.ts b/packages/ai/src/agents/think-and-prep-agent/get-think-and-prep-agent-system-prompt.ts
similarity index 96%
rename from packages/ai/src/agents/think-and-prep-agent/think-and-prep-instructions.ts
rename to packages/ai/src/agents/think-and-prep-agent/get-think-and-prep-agent-system-prompt.ts
index e27edee60..cd965f871 100644
--- a/packages/ai/src/agents/think-and-prep-agent/think-and-prep-instructions.ts
+++ b/packages/ai/src/agents/think-and-prep-agent/get-think-and-prep-agent-system-prompt.ts
@@ -1,16 +1,4 @@
-import { getPermissionedDatasets } from '@buster/access-controls';
-import type { RuntimeContext } from '@mastra/core/runtime-context';
-import type { AnalystRuntimeContext } from '../../workflows/analyst-workflow';
-import { getSqlDialectGuidance } from '../shared/sql-dialect-guidance';
-
-// Define the required template parameters
-interface ThinkAndPrepTemplateParams {
-  databaseContext: string;
-  sqlDialectGuidance: string;
-}
-
-// Template string as a function that requires parameters
-const createThinkAndPrepInstructions = (params: ThinkAndPrepTemplateParams): string => {
+export const getThinkAndPrepAgentSystemPrompt = (sqlDialectGuidance: string): string => {
   return `
 You are Buster, a specialized AI agent within an AI-powered data analyst system.
 
@@ -398,7 +386,7 @@ Once all TODO list items are addressed and submitted for review, the system will
 
 <sql_best_practices>
 - Current SQL Dialect Guidance:
-${params.sqlDialectGuidance}
+${sqlDialectGuidance}
 - Keep Queries Simple: Strive for simplicity and clarity in your SQL. Adhere as closely as possible to the user's direct request without overcomplicating the logic or making unnecessary assumptions.
 - Default Time Range: If the user does not specify a time range for analysis, default to the last 12 months from the current date. Clearly state this assumption if making it.
 - Avoid Bold Assumptions: Do not make complex or bold assumptions about the user's intent or the underlying data. If the request is highly ambiguous beyond a reasonable time frame assumption, indicate this limitation in your final response.
@@ -583,46 +571,6 @@ ${params.sqlDialectGuidance}
 Start by using the \`sequentialThinking\` to immediately start checking off items on your TODO list
 
 Today's date is ${new Date().toLocaleDateString()}.
-
----
-
-<database_context>
-${params.databaseContext}
-</database_context>
 `;
 };
 
-export const getThinkAndPrepInstructions = async ({
-  runtimeContext,
-}: { runtimeContext: RuntimeContext<AnalystRuntimeContext> }): Promise<string> => {
-  const userId = runtimeContext.get('userId');
-  const dataSourceSyntax = runtimeContext.get('dataSourceSyntax');
-
-  const datasets = await getPermissionedDatasets(userId, 0, 1000);
-
-  // Extract yml_content from each dataset and join with separators
-  const assembledYmlContent = datasets
-    .map((dataset: { ymlFile: string | null | undefined }) => dataset.ymlFile)
-    .filter((content: string | null | undefined) => content !== null && content !== undefined)
-    .join('\n---\n');
-
-  // Get dialect-specific guidance
-  const sqlDialectGuidance = getSqlDialectGuidance(dataSourceSyntax);
-
-  return createThinkAndPrepInstructions({
-    databaseContext: assembledYmlContent,
-    sqlDialectGuidance,
-  });
-};
-
-// Export the template function without dataset context for use in step files
-export const createThinkAndPrepInstructionsWithoutDatasets = (
-  sqlDialectGuidance: string
-): string => {
-  return createThinkAndPrepInstructions({
-    databaseContext: '',
-    sqlDialectGuidance,
-  })
-    .replace(/<database_context>[\s\S]*?<\/database_context>/, '')
-    .trim();
-};
diff --git a/packages/ai/src/agents/think-and-prep-agent/think-and-prep-agent.ts b/packages/ai/src/agents/think-and-prep-agent/think-and-prep-agent.ts
index 8ae534050..8ca5b8539 100644
--- a/packages/ai/src/agents/think-and-prep-agent/think-and-prep-agent.ts
+++ b/packages/ai/src/agents/think-and-prep-agent/think-and-prep-agent.ts
@@ -1,4 +1,6 @@
-import { Agent } from '@mastra/core';
+import { hasToolCall, type ModelMessage, stepCountIs, streamText } from "ai";
+import { wrapTraced } from "braintrust";
+import z from "zod";
 import {
   executeSql,
   messageUserClarifyingQuestion,
@@ -7,29 +9,79 @@ import {
   submitThoughts,
 } from '../../tools';
 import { Sonnet4 } from '../../utils/models/sonnet-4';
+import { getThinkAndPrepAgentSystemPrompt } from './get-think-and-prep-agent-system-prompt';
 
-const DEFAULT_OPTIONS = {
-  maxSteps: 18,
-  temperature: 0,
-  maxTokens: 10000,
-  providerOptions: {
-    anthropic: {
-      disableParallelToolCalls: true,
-    },
-  },
+const DEFAULT_CACHE_OPTIONS = {
+  anthropic: { cacheControl: { type: "ephemeral", ttl: "1h" } },
 };
 
-export const thinkAndPrepAgent = new Agent({
-  name: 'Think and Prep Agent',
-  instructions: '', // We control the system messages in the step at stream instantiation
-  model: Sonnet4,
-  tools: {
-    sequentialThinking,
-    executeSql,
-    respondWithoutAssetCreation,
-    submitThoughts,
-    messageUserClarifyingQuestion,
-  },
-  defaultGenerateOptions: DEFAULT_OPTIONS,
-  defaultStreamOptions: DEFAULT_OPTIONS,
+const STOP_CONDITIONS = [
+  stepCountIs(18), 
+  hasToolCall("submitThoughts"),
+  hasToolCall("respondWithoutAssetCreation"),
+  hasToolCall("messageUserClarifyingQuestion")
+];
+
+const ThinkAndPrepAgentOptionsSchema = z.object({
+  sql_dialect_guidance: z
+    .string()
+    .describe("The SQL dialect guidance for the think and prep agent."),
 });
+
+const ThinkAndPrepStreamOptionsSchema = z.object({
+  messages: z
+    .array(z.custom<ModelMessage>())
+    .describe("The messages to send to the think and prep agent."),
+});
+
+export type ThinkAndPrepAgentOptionsSchema = z.infer<
+  typeof ThinkAndPrepAgentOptionsSchema
+>;
+export type ThinkAndPrepStreamOptions = z.infer<typeof ThinkAndPrepStreamOptionsSchema>;
+
+export function createThinkAndPrepAgent(
+  thinkAndPrepAgentSchema: ThinkAndPrepAgentOptionsSchema,
+) {
+  const steps: never[] = [];
+
+  const systemMessage = {
+    role: "system",
+    content: getThinkAndPrepAgentSystemPrompt(
+      thinkAndPrepAgentSchema.sql_dialect_guidance,
+    ),
+    providerOptions: DEFAULT_CACHE_OPTIONS,
+  } as ModelMessage;
+
+  async function stream({ messages }: ThinkAndPrepStreamOptions) {
+    return wrapTraced(
+      () =>
+        streamText({
+          model: Sonnet4,
+          tools: {
+            sequentialThinking,
+            executeSql,
+            respondWithoutAssetCreation,
+            submitThoughts,
+            messageUserClarifyingQuestion,
+          },
+          messages: [systemMessage, ...messages],
+          stopWhen: STOP_CONDITIONS,
+          toolChoice: "required",
+          maxOutputTokens: 10000,
+          temperature: 0,
+        }),
+      {
+        name: "Think and Prep Agent",
+      },
+    )();
+  }
+
+  async function getSteps() {
+    return steps;
+  }
+
+  return {
+    stream,
+    getSteps,
+  };
+}
diff --git a/packages/ai/src/steps/think-and-prep-step.ts b/packages/ai/src/steps/think-and-prep-step.ts
index 8b8d112ce..1f6595910 100644
--- a/packages/ai/src/steps/think-and-prep-step.ts
+++ b/packages/ai/src/steps/think-and-prep-step.ts
@@ -6,8 +6,8 @@ import type { CoreMessage } from 'ai';
 import { wrapTraced } from 'braintrust';
 import { z } from 'zod';
 import { getSqlDialectGuidance } from '../agents/shared/sql-dialect-guidance';
-import { thinkAndPrepAgent } from '../agents/think-and-prep-agent/think-and-prep-agent';
-import { createThinkAndPrepInstructionsWithoutDatasets } from '../agents/think-and-prep-agent/think-and-prep-instructions';
+import { createThinkAndPrepAgent } from '../agents/think-and-prep-agent/think-and-prep-agent';
+import { getThinkAndPrepAgentSystemPrompt } from '../agents/think-and-prep-agent/get-think-and-prep-agent-system-prompt';
 import type { thinkAndPrepWorkflowInputSchema } from '../schemas/workflow-schemas';
 import { ChunkProcessor } from '../utils/database/chunk-processor';
 import {
@@ -248,60 +248,48 @@ ${databaseContext}
           ),
         });
 
+        // Create the agent instance
+        const thinkAndPrepAgent = createThinkAndPrepAgent({
+          sql_dialect_guidance: sqlDialectGuidance,
+        });
+
         const wrappedStream = wrapTraced(
           async () => {
-            // Create system messages with dataset context and instructions
-            const systemMessages: CoreMessage[] = [
-              {
-                role: 'system',
-                content: createThinkAndPrepInstructionsWithoutDatasets(sqlDialectGuidance),
-                providerOptions: DEFAULT_CACHE_OPTIONS,
-              },
-              {
-                role: 'system',
-                content: createDatasetSystemMessage(assembledYmlContent),
-                providerOptions: DEFAULT_CACHE_OPTIONS,
-              },
-            ];
+            // Create dataset system message
+            const datasetSystemMessage: CoreMessage = {
+              role: 'system',
+              content: createDatasetSystemMessage(assembledYmlContent),
+              providerOptions: DEFAULT_CACHE_OPTIONS,
+            };
 
-            // Combine system messages with conversation messages
-            const messagesWithSystem = [...systemMessages, ...messages];
+            // Combine dataset system message with conversation messages
+            const messagesWithDataset = [datasetSystemMessage, ...messages];
 
-            // Create stream directly without retryableAgentStreamWithHealing
-            const stream = await thinkAndPrepAgent.stream(messagesWithSystem, {
-              toolCallStreaming: true,
-              runtimeContext,
-              maxRetries: 5,
-              abortSignal: abortController.signal,
-              toolChoice: 'required',
-              onChunk: createOnChunkHandler({
-                chunkProcessor,
-                abortController,
-                finishingToolNames: [
-                  'submitThoughts',
-                  'respondWithoutAssetCreation',
-                  'messageUserClarifyingQuestion',
-                ],
-                onFinishingTool: () => {
-                  // Set finished = true for respondWithoutAssetCreation and messageUserClarifyingQuestion
-                  // submitThoughts should abort but not finish so workflow can continue
-                  const finishingToolName = chunkProcessor.getFinishingToolName();
-                  if (
-                    finishingToolName === 'respondWithoutAssetCreation' ||
-                    finishingToolName === 'messageUserClarifyingQuestion'
-                  ) {
-                    finished = true;
-                  }
-                },
-              }),
-              onError: createRetryOnErrorHandler({
-                retryCount,
-                maxRetries,
-                workflowContext: {
-                  currentStep: 'think-and-prep',
-                  availableTools,
-                },
-              }),
+            // Create stream using the new agent pattern
+            const stream = await thinkAndPrepAgent.stream({
+              messages: messagesWithDataset,
+            });
+
+            // Handle streaming with chunk processor
+            stream.onChunk = createOnChunkHandler({
+              chunkProcessor,
+              abortController,
+              finishingToolNames: [
+                'submitThoughts',
+                'respondWithoutAssetCreation',
+                'messageUserClarifyingQuestion',
+              ],
+              onFinishingTool: () => {
+                // Set finished = true for respondWithoutAssetCreation and messageUserClarifyingQuestion
+                // submitThoughts should abort but not finish so workflow can continue
+                const finishingToolName = chunkProcessor.getFinishingToolName();
+                if (
+                  finishingToolName === 'respondWithoutAssetCreation' ||
+                  finishingToolName === 'messageUserClarifyingQuestion'
+                ) {
+                  finished = true;
+                }
+              },
             });
 
             return stream;
diff --git a/packages/ai/src/utils/memory/ai-sdk-bundling.test.ts b/packages/ai/src/utils/memory/ai-sdk-bundling.test.ts
deleted file mode 100644
index 7a23254c2..000000000
--- a/packages/ai/src/utils/memory/ai-sdk-bundling.test.ts
+++ /dev/null
@@ -1,279 +0,0 @@
-import type { CoreMessage } from 'ai';
-import { describe, expect, test } from 'vitest';
-import { validateArrayAccess } from '../validation-helpers';
-import { extractMessageHistory } from './message-history';
-
-describe('AI SDK Message Bundling Issues', () => {
-  test('identify when AI SDK returns bundled messages', () => {
-    // The AI SDK tends to bundle multiple tool calls in a single assistant message
-    // when parallel tool calls are made, even with disableParallelToolCalls
-    const aiSdkResponse: CoreMessage[] = [
-      {
-        role: 'user',
-        content: 'Analyze our customer data',
-      },
-      {
-        role: 'assistant',
-        content: [
-          {
-            type: 'tool-call',
-            toolCallId: 'call_ABC123',
-            toolName: 'sequentialThinking',
-            args: { thought: 'First, I need to understand the data structure' },
-          },
-          {
-            type: 'tool-call',
-            toolCallId: 'call_DEF456',
-            toolName: 'executeSql',
-            args: { statements: ['SELECT COUNT(*) FROM customers'] },
-          },
-          {
-            type: 'tool-call',
-            toolCallId: 'call_GHI789',
-            toolName: 'submitThoughts',
-            args: {},
-          },
-        ],
-      },
-      {
-        role: 'tool',
-        content: [
-          {
-            type: 'tool-result',
-            toolCallId: 'call_ABC123',
-            toolName: 'sequentialThinking',
-            result: { success: true },
-          },
-        ],
-      },
-      {
-        role: 'tool',
-        content: [
-          {
-            type: 'tool-result',
-            toolCallId: 'call_DEF456',
-            toolName: 'executeSql',
-            result: { results: [{ count: 100 }] },
-          },
-        ],
-      },
-      {
-        role: 'tool',
-        content: [
-          {
-            type: 'tool-result',
-            toolCallId: 'call_GHI789',
-            toolName: 'submitThoughts',
-            result: {},
-          },
-        ],
-      },
-    ];
-
-    // Our extraction should fix this
-    const fixed = extractMessageHistory(aiSdkResponse);
-
-    // Should be properly interleaved now
-    expect(fixed).toHaveLength(7); // user + 3*(assistant + tool)
-
-    // Check the pattern
-    const msg0 = validateArrayAccess(fixed, 0, 'fixed messages');
-    const msg1 = validateArrayAccess(fixed, 1, 'fixed messages');
-    const msg2 = validateArrayAccess(fixed, 2, 'fixed messages');
-    const msg3 = validateArrayAccess(fixed, 3, 'fixed messages');
-    const msg4 = validateArrayAccess(fixed, 4, 'fixed messages');
-    const msg5 = validateArrayAccess(fixed, 5, 'fixed messages');
-    const msg6 = validateArrayAccess(fixed, 6, 'fixed messages');
-
-    expect(msg0.role).toBe('user');
-    expect(msg1.role).toBe('assistant');
-    if (msg1.role === 'assistant' && Array.isArray(msg1.content)) {
-      const content = validateArrayAccess(msg1.content, 0, 'assistant content');
-      if ('toolCallId' in content) {
-        expect(content.toolCallId).toBe('call_ABC123');
-      }
-    }
-    expect(msg2.role).toBe('tool');
-    if (msg2.role === 'tool' && Array.isArray(msg2.content)) {
-      const content = validateArrayAccess(msg2.content, 0, 'tool content');
-      if ('toolCallId' in content) {
-        expect(content.toolCallId).toBe('call_ABC123');
-      }
-    }
-    expect(msg3.role).toBe('assistant');
-    if (msg3.role === 'assistant' && Array.isArray(msg3.content)) {
-      const content = validateArrayAccess(msg3.content, 0, 'assistant content');
-      if ('toolCallId' in content) {
-        expect(content.toolCallId).toBe('call_DEF456');
-      }
-    }
-    expect(msg4.role).toBe('tool');
-    if (msg4.role === 'tool' && Array.isArray(msg4.content)) {
-      const content = validateArrayAccess(msg4.content, 0, 'tool content');
-      if ('toolCallId' in content) {
-        expect(content.toolCallId).toBe('call_DEF456');
-      }
-    }
-    expect(msg5.role).toBe('assistant');
-    if (msg5.role === 'assistant' && Array.isArray(msg5.content)) {
-      const content = validateArrayAccess(msg5.content, 0, 'assistant content');
-      if ('toolCallId' in content) {
-        expect(content.toolCallId).toBe('call_GHI789');
-      }
-    }
-    expect(msg6.role).toBe('tool');
-    if (msg6.role === 'tool' && Array.isArray(msg6.content)) {
-      const content = validateArrayAccess(msg6.content, 0, 'tool content');
-      if ('toolCallId' in content) {
-        expect(content.toolCallId).toBe('call_GHI789');
-      }
-    }
-  });
-
-  test('handle case where AI SDK partially bundles messages', () => {
-    // Sometimes the AI SDK might bundle some calls but not others
-    const partiallyBundled: CoreMessage[] = [
-      {
-        role: 'user',
-        content: 'Test',
-      },
-      {
-        role: 'assistant',
-        content: [
-          {
-            type: 'tool-call',
-            toolCallId: 'id1',
-            toolName: 'tool1',
-            args: {},
-          },
-        ],
-      },
-      {
-        role: 'tool',
-        content: [
-          {
-            type: 'tool-result',
-            toolCallId: 'id1',
-            toolName: 'tool1',
-            result: {},
-          },
-        ],
-      },
-      {
-        role: 'assistant',
-        content: [
-          {
-            type: 'tool-call',
-            toolCallId: 'id2',
-            toolName: 'tool2',
-            args: {},
-          },
-          {
-            type: 'tool-call',
-            toolCallId: 'id3',
-            toolName: 'tool3',
-            args: {},
-          },
-        ],
-      },
-      {
-        role: 'tool',
-        content: [
-          {
-            type: 'tool-result',
-            toolCallId: 'id2',
-            toolName: 'tool2',
-            result: {},
-          },
-        ],
-      },
-      {
-        role: 'tool',
-        content: [
-          {
-            type: 'tool-result',
-            toolCallId: 'id3',
-            toolName: 'tool3',
-            result: {},
-          },
-        ],
-      },
-    ];
-
-    const fixed = extractMessageHistory(partiallyBundled);
-
-    // Should fix only the bundled part
-    expect(fixed).toHaveLength(7);
-
-    // First part should remain unchanged
-    const fixedMsg0 = validateArrayAccess(fixed, 0, 'fixed messages');
-    const fixedMsg1 = validateArrayAccess(fixed, 1, 'fixed messages');
-    const fixedMsg2 = validateArrayAccess(fixed, 2, 'fixed messages');
-    const fixedMsg3 = validateArrayAccess(fixed, 3, 'fixed messages');
-    const fixedMsg4 = validateArrayAccess(fixed, 4, 'fixed messages');
-    const fixedMsg5 = validateArrayAccess(fixed, 5, 'fixed messages');
-    const fixedMsg6 = validateArrayAccess(fixed, 6, 'fixed messages');
-
-    const partialMsg0 = validateArrayAccess(partiallyBundled, 0, 'partially bundled messages');
-    const partialMsg1 = validateArrayAccess(partiallyBundled, 1, 'partially bundled messages');
-    const partialMsg2 = validateArrayAccess(partiallyBundled, 2, 'partially bundled messages');
-
-    expect(fixedMsg0).toEqual(partialMsg0);
-    expect(fixedMsg1).toEqual(partialMsg1);
-    expect(fixedMsg2).toEqual(partialMsg2);
-
-    // Second part should be unbundled
-    if (fixedMsg3.role === 'assistant' && Array.isArray(fixedMsg3.content)) {
-      const content = validateArrayAccess(fixedMsg3.content, 0, 'assistant content');
-      if ('toolCallId' in content) {
-        expect(content.toolCallId).toBe('id2');
-      }
-    }
-    if (fixedMsg4.role === 'tool' && Array.isArray(fixedMsg4.content)) {
-      const content = validateArrayAccess(fixedMsg4.content, 0, 'tool content');
-      if ('toolCallId' in content) {
-        expect(content.toolCallId).toBe('id2');
-      }
-    }
-    if (fixedMsg5.role === 'assistant' && Array.isArray(fixedMsg5.content)) {
-      const content = validateArrayAccess(fixedMsg5.content, 0, 'assistant content');
-      if ('toolCallId' in content) {
-        expect(content.toolCallId).toBe('id3');
-      }
-    }
-    if (fixedMsg6.role === 'tool' && Array.isArray(fixedMsg6.content)) {
-      const content = validateArrayAccess(fixedMsg6.content, 0, 'tool content');
-      if ('toolCallId' in content) {
-        expect(content.toolCallId).toBe('id3');
-      }
-    }
-  });
-
-  test('verify already correct messages pass through unchanged', () => {
-    const correctlyFormatted: CoreMessage[] = [
-      { role: 'user', content: 'Test' },
-      {
-        role: 'assistant',
-        content: [{ type: 'tool-call', toolCallId: 'id1', toolName: 'tool1', args: {} }],
-      },
-      {
-        role: 'tool',
-        content: [{ type: 'tool-result', toolCallId: 'id1', toolName: 'tool1', result: {} }],
-      },
-      {
-        role: 'assistant',
-        content: [{ type: 'tool-call', toolCallId: 'id2', toolName: 'tool2', args: {} }],
-      },
-      {
-        role: 'tool',
-        content: [{ type: 'tool-result', toolCallId: 'id2', toolName: 'tool2', result: {} }],
-      },
-    ];
-
-    const result = extractMessageHistory(correctlyFormatted);
-
-    // Should be unchanged
-    expect(result).toEqual(correctlyFormatted);
-    expect(result).toHaveLength(5);
-  });
-});