Goal Setting & Monitoring

Key points

• A vague goal makes an agent drift and waste money.
• Success criteria = a checklist the agent (and you) can verify.
• Monitoring = checking progress each loop, plus stopping conditions.

What 'Goal Setting & Monitoring' means

Goal Setting is giving the agent a specific, measurable target plus clear success criteria (how we know it's finished). Monitoring is continuously comparing the agent's progress against that target and applying stopping conditions so it doesn't loop forever or wander off task.

Note: Clear goal + measurable success criteria + progress checks + stop rules = an agent that finishes the RIGHT job.

Vague goal vs measurable goal

VAGUE GOAL (agent drifts) MEASURABLE GOAL (agent focuses) ────────────────────────── ───────────────────────────────

"Make the report better" "Cut the report to <500 words, │ add a 3-bullet summary, fix ▼ all spelling." ┌─────────────┐ │ │ agent guesses│ ▼ │ endlessly... │ ┌──────────────┐ │ 🌀 🌀 🌀 │ │ check 3 boxes │ └─────────────┘ │ [x] <500 words│ │ │ [x] summary │ ▼ │ [x] spelling │ ❓ never sure it's done └──────┬───────┘ ▼ ✅ all 3 met = DONE

The 4 ingredients of a good goal

• Specific target — One clear outcome, not a fuzzy wish. Example: "Book 1 flight under ₹5000 to Delhi on Friday" not "sort out travel".
• Success criteria — A checklist that is true/false, so 'done' is not an opinion. Example: price < 5000 AND date == Friday AND seat == confirmed.
• Progress signal — Something you can measure each loop to see if it's improving. Example: criteria_met = 1 of 3 → 2 of 3 → 3 of 3.
• Stopping conditions — Rules that end the run: success, max steps, budget, or no progress. Example: Stop when all criteria met OR after 10 steps OR cost > ₹20.

The monitoring loop (goal in the middle)

┌─────────────────────────┐ │ GOAL + SUCCESS CHECK │ │ [ ] <500 words │ │ [ ] has 3-bullet summary│ │ [ ] spelling clean │ └────────────┬────────────┘ │ compare every step ┌─────────────────────────┼─────────────────────────┐ ▼ ▼ ▼ ┌─────────┐ act ┌─────────────────┐ observe ┌──────────────┐ │ THINK │───────► │ DO ONE STEP │─────────► │ CHECK PROGRESS│ └─────────┘ └─────────────────┘ └──────┬───────┘ ▲ │ │ not all boxes ticked AND steps left AND budget │ └────────────────────────────────────────────────────┘ │ all boxes ticked │ OR out of steps/budget ▼ 🛑 STOP

A tiny code example (read it like English)

Here the goal is a checklist of functions. Each loop we re-run the checklist, count how many pass, and stop the moment all pass or we hit a safety limit.

MAX_STEPS = 10
MAX_COST = 20.0  # rupees

def success_criteria(report):
    return {
        "under_500_words": len(report.split()) < 500,
        "has_summary": "Summary:" in report,
        "spelling_ok": spell_check(report) == 0,
    }

report, cost, step = draft(), 0.0, 0
while True:
    checks = success_criteria(report)
    met = sum(checks.values())
    print(f"step={step} progress={met}/3 cost=₹{cost:.1f}")  # MONITOR

    if all(checks.values()):
        print("✅ goal reached"); break              # stop: success
    if step >= MAX_STEPS or cost >= MAX_COST:
        print("🛑 stopped: hit a limit"); break      # stop: safety

    report, spent = improve(report, failing=checks)  # ACT
    cost += spent; step += 1

When to invest in goals & monitoring

Goal & monitoring mistakes

Remember these lines

• If you can't write 'done' as a checklist, the goal is too vague.
• Every loop: measure progress, then decide whether to continue or stop.
• Three stop rules to always have: success, max-steps, max-budget.

Key takeaways

• Goal Setting gives the agent a specific target plus measurable success criteria.
• Monitoring compares progress to that target every step of the loop.
• Stopping conditions (success, max-steps, max-budget, no-progress) prevent runaway agents.
• If you can't express 'done' as true/false checks, your goal is still too vague.

Frequently Asked Questions

Related topics

• Learning & Adaptation
• Memory Management Pattern
• Multi-Agent Collaboration
• Prompt Chaining Pattern
• ReAct Pattern (Reason + Act)

Browse all Agentic AI Patterns topics →