Instant, Reasoning, and Agent, three modes, one coach

Most AI coaches give you one setting: you ask, it answers. Atomic Metrix gives you three. Each one trades speed for depth, and knowing which to reach for is the difference between a quick lookup and a fully rescheduled training week.

This guide walks through Instant, Reasoning, and Agent mode: what each one can do, what it costs, and when to use which. No jargon, no model names, just the practical playbook.

Why three modes instead of one?

Think of it like gears on your bike. You could ride everything in one gear, but you'd grind on climbs and spin out on descents. AI modes work the same way:

Instant is your small ring, fast cadence, low effort. It draws on general knowledge and answers in seconds. No training data needed, no quota burned.

Reasoning is your big ring on the flats, it reads your actual training data (fitness, fatigue, power zones, recent rides) and gives you a personalized analysis. Takes a few more seconds, uses more quota, but the answer is about you.

Agent is your full drivetrain. It can read your data and take action: build a training plan, reschedule workouts, mark sessions complete. It's the only mode that changes things, and that's why it's the most powerful, and the most expensive.

Use the lightest mode that gets the job done. Your monthly quota increases as capability decreases,100 Instant, 20 Reasoning, 10 Agent.

What each mode can actually do

Tap each mode below to see its capabilities, speed, data access, and monthly quota. This is the reference card, bookmark it.

The capability spectrum

The three modes sit on a spectrum from fast and general to slow and personal. Here's how to think about it:

Instant is stateless, it doesn't know who you are. It's a well-read training textbook that happens to respond in natural language. Ask it "What heart rate zone is best for long rides?" and you'll get a solid, accurate answer. Ask it "Am I overtraining?" and you'll get generic signs to watch for, but nothing about your actual training load.

Reasoning is contextual, it can look up your training status, pull your recent activities, check your power zones, and read your fitness/fatigue balance. It calls internal tools (like get_training_status or get_recent_activities) to fetch real numbers, then reasons over them. The answer you get is grounded in your data, not textbook averages.

Agent is agentic, it can do everything Reasoning does, plus take multi-step actions. It can generate a training plan, schedule workouts onto your calendar, modify an existing plan when life gets in the way, and mark workouts as complete. If the task involves changing something, you need Agent.

Same question, three different answers

The best way to understand the difference is to see it. Pick a question below and toggle between modes, watch how the depth, data usage, and action-taking ability change with each step up.

Notice the pattern: Instant gives you knowledge, Reasoning gives you insight, and Agent gives you execution. The tool calls in the Reasoning and Agent responses aren't decoration, they represent real queries against your training database.

The decision framework

You don't need to memorize capabilities. Instead, ask yourself two questions:

1. Does this need my personal data? If no → Instant. If yes → keep going.

2. Does this need to change something? If no → Reasoning. If yes → Agent.

That's really it. "What is FTP?" → Instant. "What's my FTP trend?" → Reasoning. "Build me a plan based on my FTP" → Agent.

Which mode should I use?

Still not sure? Tap what you want to do, and we'll tell you which mode fits and why.

Making the most of your quota

Each mode has a monthly message limit: 100 Instant, 20 Reasoning, 10 Agent. This isn't arbitrary, it reflects the computational cost of each mode. Here's how to stretch them:

Batch your Agent requests. Instead of asking Agent to "add a ride for tomorrow" then "add one for Wednesday" then "add one for Friday," say "Schedule my three key sessions for this week." One Agent message can do what three would.

Use Reasoning as a scout. Before asking Agent to build a plan, ask Reasoning "What's my current form and what should my next training block look like?" The insight helps you give Agent a more specific, better-targeted request, which means a better plan on the first try.

Don't waste Agent on lookups. "What's my CTL?" is a Reasoning question. "How has my CTL changed this month?" is a Reasoning question. Save Agent for when you need action, not answers.

Real-world workflow examples

Here's how a typical training week might use all three modes:

Monday morning: You're curious whether a rest day is warranted. Ask Reasoning: "Am I recovered enough for intervals today?" It checks your TSB, yesterday's TSS, and your recent load trend. Cost: 1 Reasoning message.

Monday afternoon: Reasoning said to take it easy. You want to know what a good recovery ride looks like. Ask Instant: "What intensity and duration for an active recovery ride?" Cost: 1 Instant message.

Wednesday: You realize your race is 3 weeks out and you haven't planned a taper. Ask Agent: "Build a 3-week taper for my road race on May 10th." Agent checks your current fitness, generates a periodized taper plan, and schedules the workouts. Cost: 1 Agent message.

Friday: You missed Thursday's session because of rain. Ask Agent: "I missed yesterday, adjust my plan for the rest of this week." Agent reads what you missed, shifts the key session, and updates your calendar. Cost: 1 Agent message.

Total weekly cost: 2 Instant, 1 Reasoning, 2 Agent. Well within your monthly budget, and every message was matched to the right mode.

The bottom line

Three modes isn't complexity for its own sake. It's the same principle behind gearing: match the tool to the terrain. Instant for knowledge, Reasoning for insight, Agent for action.

Start with Instant. Upgrade to Reasoning when you need your numbers. Reach for Agent when you need something done. That simple framework will carry you through 95% of conversations with your AI coach.