AI Coach Cloud. Claude + Intervals.icu + agentic workout posting and manipulation. Looking for founding members to join, use, and give feedback

TsSource · 15 May 2026 00:03

About me:
Hi, I’m Rohit. Former paramedic for nearly a decade, endurance training junkie, and multiple 70.3 distance triathlon finisher. In a nutshell, my life revolves around getting my training in, figuring out how I feel, deciding on when to push and when to hold back, and reading my own data. Having been coached by a human coach for many years, I wanted a coach I could access 24/7, intimately customized to me and my data, and a coach that follows me and not the other way around. So I built Sharma AI Coach.

What it is:
-Demo Video (~2min): https://youtu.be/t94TFathyCs

A hosted SaaS that:

Reads your Intervals.icu data — CTL, ATL, form, ramp rate, planned + completed workouts, HRV, sleep, fatigue
Holds a coaching conversation about it
Writes structured workouts directly back to your ICU calendar with zones, intervals, and the workout-text syntax Intervals.icu parses natively (so Zwift/Garmin/Wahoo sync downstream is automatic)
The AI model can also completely manipulate your workout: moving, replacing, deleting, modifying, etc..
The calendar section also allows for drag and drop movement of workouts, which the AI model also sees, so your next prompt will be up to date on the specific manipulation that occurred.
All updates and workout posts get pushed to your intervals.icu calendar.
Pushes back when your ramp is too aggressive instead of rubber-stamping the plan
Runs on Anthropic’s Claude (Sonnet 4.6 for daily coaching, Opus 4.7 for race plans)

Pricing:

Looking for founding members for a limited time, to use the service and give me constructive feedback on what to add and/or improve.

Founding cohort: $14.99/mo, locked for life-of-subscription.
Monthly: $24.99/mo
Annual: $249/yr (~$20.75/mo effective)
7-day free trial on every tier
Stripe billing, cancel anytime through the customer portal

Try it out:

I look forward to improving my service with your feedback.

TsSource · 16 May 2026 04:30

Quick update for anyone following.

Shipped a stack of meaningful improvements this week from real production use, thanks to the first wave of founding-member traffic this week thus far.

Agentic calendar-write hardening. Two related fixes for multi-week bulk-plan writes:

Server-side date arithmetic. The AI model is right MOST of the time, but an edge case brought up a date-related hallucination. Programmatic fix implemented. AI model absolved of date arithmetic, where hallucination chances compound with bulk posting workouts.
Transparent rate-limit handling. Bulk-writes tripped backend API rate limit in edge cases. Now the model throttles bulk actions at the orchestrator level with max 3 retries per request. Failures now resolve gracefully with transparent language on why.

Power-zone rendering on Ride / VirtualRide. Intensity zone bars now display correctly on indoor and virtual rides where they previously didn’t.

Workout duration in coaching context. Fixed a quiet bug where activity durations weren’t being passed into the coach’s view of recent training. The coach now sees the full picture (distance + duration + load) on every activity it analyzes.

I would greatly appreciate my current founding members and future founding members (still for a limited time) to use my product and give me constructive feedback on bugs you may find, and which areas to improve.

This is a passion project of mine, and I use it myself every day. But these edge cases would’ve never surfaced from my use alone. So I thank you all, who have given me feedback, for your support, patience, and time.

TsSource · 20 May 2026 02:02

Update: New features and upgrades added to AI Coach Cloud:

-> Desktop month calendar view (alongside week + day views)

→ New Predictions page: race-time predictions across 5K / 10K /
half-marathon / marathon (run) and 40K / 100K (bike), with PR
detection from your activity history. Predictions are calibrated
to your current CTL and shown as low/mid/high ranges rather than
single-point guesses. Each prediction shows the source activity
it drew from, so you can see why the number is what it is.

Happy to take feature requests or answer questions.

Try it out at Sharma Automation | AI Coach for Endurance Athletes

TsSource · 25 May 2026 15:57

New feature: daily briefings.

AI Coach Cloud now checks in every morning. It reads your last few days of training, looks at how your fitness and fatigue are tracking against your plan, and writes you a short plain-English briefing for the day.

If it spots something worth changing: added fatigue, a missed session, a week that’s drifted off plan, it drafts the revised calendar-ready workouts. You review and accept or dismiss in one click. Nothing lands on your calendar unless you approve it, and it only touches the sessions it’s managing; anything you built yourself is left alone.

Thomas_Poppe · 1 June 2026 06:54

Hi,

Been using it a little bit the last couple of days.
Prefer how this tool keeps intervals.icu as the source of truth (in terms of calendar and values) and doesn’t attempt to do a sync. Also a clear data privacy policy, which is missing from many other tools on this forum.

There are a few issues that I’ve encountered.

Cannot delete workouts through AI Coach (it notices it fails, and suggests to do it manually.)
Gives me race predications for my Half Distance Triathlon, but those are founded on the wrong distances. It’s a 90KM bike and 2KM swim (the event details contains this information correctly)
Run workouts do not include the “PACE” or “HR” in the workout description, so they don’t show up in intervals.icu with a graph.

I’ll do some further testing in the coming weeks.
Anyhow nice job so far.

TsSource · 1 June 2026 23:23

Hi Thomas,

Two of your three points are now live:

You can now delete workouts with a 5-minute undo if you change your mind.
Runs now come with prescribed pace and HR targets (if set within Intervals) in both intervals.icu and the AI Coach Cloud app. HR sessions render a zone graph in intervals (if HR is set in intervals), and the AI Coach Cloud workout drawer (upon clicking on the workout pill in the calendar) shows the full pace and HR zone breakdown. Pacing and their graphs gracefully fallback if those values are not set in intervals. AI Couch Cloud uses your stored threshold pace on the AI Coach Cloud platform to compute proper pacing with each workout step, which then gets pushed to Intervals (without the graph if threshold pace isn’t set in intervals). HR has no fallback at the moment.

Your third item, the triathlon race predictions using proper per-discipline distances (swim / bike / run), is what I’m building next.

Thanks for the sharp feedback! It’s shaping the product directly. I will keep you updated on the race predictions adjustment. More soon.

Please try out the 2 live implementations for yourself. Note: newly created runs by the AI Coach will have the proper pacing/hr syntax.

-Rohit

Thomas_Poppe · 3 June 2026 16:35

Thanks for the update and the effort out in the app.

First attempt didn’t work out though. It combined the HR and PACE, which cannot be done in the workout builder. It must be one of either.

TsSource · 4 June 2026 02:26

Thank you.

Every newly created run now uses a single target type, so it renders cleanly in the intervals editor. I also re-processed all mixed-type runs, so they should be either all hr or pace-based, depending on which criteria is met (easy - aerobic: hr. any kind of structure-intensity: pace)

Model-built easy, aerobic, and recovery runs are heart-rate-based (a % of your threshold HR). Anything with structure or intensity (intervals, tempo, threshold, VO2) is pace-based.

On the race predictions from your earlier note: a race now carries its full discipline (triathlon, duathlon, run, bike, swim, etc.) with each leg’s distance, instead of a single type and distance. Predictions are leg-by-leg and sport-specific.

Each leg gets its own projected time where there’s enough history, and anything that can’t be called honestly is flagged rather than guessed.

When you add a race, pick the discipline and enter the per-leg distances; the leg predictions then show on the calendar, the day view, and the predictions page.

Please try creating another run on AI Coach Cloud. It should now render correctly on the app and on Intervals. Also take a look at race predictions. If a race is created on Intervals side first, it will appear on AI Coach Cloud, where you will be nudged to enter the specific discipline (triathlon) and its legs with specific distances within settings. Once saved/created/updated within settings, the predictions will recompute.

Predictions automatically recompute every 24 hours inline with your training data.

I appreciate the detailed feedback. Please let me know how it goes on your end, or if you find anything else, or any feature requests.