diff --git a/agents/tech-trends-agent.json b/agents/tech-trends-agent.json
index 502f138..9d6b181 100644
--- a/agents/tech-trends-agent.json
+++ b/agents/tech-trends-agent.json
@@ -1,13 +1,19 @@
 {
   "agent_name": "tech-trends-agent",
+  "phase": "1",
   "definition": {
     "model": "${GPT_DEPLOYMENT}",
     "instructions_file": "prompts/tech-trends-agent.md",
-    "tools": []
+    "tools": [
+      { "type": "web_search" }
+    ]
   },
   "eval": {
     "dataset": "evals/golden-dataset.json",
-    "phase_filter": null,
+    "phase_filter": "1",
     "config": "evals/eval-config.json"
-  }
+  },
+  "_model_history": [
+    { "model": "gpt-4o-2024-11-20", "from": "2025-01-10", "to": null, "reason": "initial" }
+  ]
 }
diff --git a/evals/eval-config.json b/evals/eval-config.json
index 832031e..5cafb0a 100644
--- a/evals/eval-config.json
+++ b/evals/eval-config.json
@@ -1,9 +1,9 @@
 {
   "evaluators": [
-    "TaskAdherenceEvaluator",
-    "RelevanceEvaluator",
-    "GroundednessEvaluator",
-    "CoherenceEvaluator"
+    "builtin.task_adherence",
+    "builtin.relevance",
+    "builtin.groundedness",
+    "builtin.coherence"
   ],
   "thresholds": {
     "task_adherence": 0.80,
@@ -11,6 +11,6 @@
     "groundedness": 0.75,
     "coherence": 0.80
   },
-  "phase_filter": null,
-  "notes": "Phase 2: All queries evaluated — both web search (Phase 1) and data analysis (Phase 2)."
+  "phase_filter": "1",
+  "notes": "Phase 1: Only web search queries evaluated. Phase 2 data analysis queries excluded."
 }
diff --git a/prompts/tech-trends-agent.md b/prompts/tech-trends-agent.md
index da79c14..e01c7e9 100644
--- a/prompts/tech-trends-agent.md
+++ b/prompts/tech-trends-agent.md
@@ -27,15 +27,3 @@ Always structure responses as:
 ## Tone
 Professional, objective, and jargon-aware. Assume the user is a technology
 professional who does not need basic concepts explained.
-
-## Data Analysis (Phase 2)
-You now have access to a code interpreter. Use it when:
-- The user asks you to calculate, compare, or rank numerical data
-- You have retrieved structured data (tables, CSVs) and analysis would add value
-- You need to produce a formatted comparison table from raw information
-
-When using code interpreter:
-1. First retrieve the data via web search
-2. Then write and run Python code to process or compare it
-3. Present results with the code output clearly labelled
-4. Always show the source of the raw data alongside the computed result