One of the AI projects I’ve been working on lately is Hermes profiles. Hermes is a great little personal-assistant-style agent harness - it can do coding, but that’s not its niche. You can create a profile for different tasks and responsibilities, so I recently created one that focuses on being a personal trainer and a nutritionist. It helps me plan my workouts and plan my meals.

The Personal Trainer

It’s hooked up to my TrainingPeaks account, so it can see all of the data from my workouts, like training load, how exhausted I am, and all of those things. It can add notes about how I felt about a workout. And it knows what my training goal is, so based on my current workload it can tell me whether I need to back off on my training intensity or if it’s time to ramp up.

# Daily training and health check

The load picture is generally stable rather than escalating: TrainingPeaks CTL is 11.8, ATL 12.4,
and TSB −2.4 today; over the last two weeks it rose briefly to ATL 18 before easing. Recent
logged training includes three runs (Oct 5, 7, 12), walks, yoga, and strength on Oct 14. Today’s
planned session is an easy 20–30 minute walk or yoga.

Recovery signals are mixed, so I’d keep today genuinely easy and avoid adding a run. TrainingPeaks
sleep over Oct 1–14 averaged about 7.3 hours, with the latest recorded night at 5.7 hours; HRV was
41 on Oct 14, within a variable recent range of 36–51. Pulse was 56 on Oct 14 and 55 today. No
workout-feeling or pain report is available, so check in with how your Achilles, back, and energy
feel before the next run; keep the familiar gentle warm-up and cool-down.

That is excellent feedback and exactly the kind of information I need to not overwork myself in my training!

The Nutritionist

On the nutritionist side, I had it start doing meal planning. I gave it all my dietary preferences: I want something that’s going to be high protein, high fiber, and low sugar. Then I had it prioritize some other metrics. I want meals that are:

  • Easy to make
  • Fast to make
  • Cheap
  • Tasty
  • Scales from 2-4 people

Then I told it to give me three recipes once a week and schedule them on specific days. It adds those recipes to the personal knowledge base that it manages, then puts them on my calendar, so I have a link to the recipe and everything. It also gives me a breakdown of the cost and estimated nutrition for each recipe.

# Banza Chicken Marinara with Spinach and Broccoli

A protein-forward pasta dinner using chickpea pasta, chicken, tomato sauce, and plenty of vegetables.
Quantities are for 2 / 4 servings.

## Ingredients
- Dry Banza or other chickpea pasta, 4 oz / 8 oz
- Chicken, 1/2 lb / 1 lb, cut into bite-size pieces
- Marinara, 1 cup / 2 cups
- Spinach, 2 cups / 4 cups
- Broccoli, 1 small head / 2 small heads, cut into florets
- Optional: a modest sprinkle of cheese
- Salt, black pepper, and Italian herbs, to taste

## Method
1. Cook pasta according to package directions; drain.
2. Meanwhile, saute or roast the chicken until cooked through (165 F at the thickest part).
3. Warm the marinara, stir in the chicken and spinach, and cook until the spinach wilts.
4. Toss with the pasta and serve with steamed or roasted broccoli. Add a modest sprinkle of cheese if desired.

## Scaling and prep
Cook pasta close to serving time for best texture. Leftover chicken and sauce can be refrigerated
promptly and reheated thoroughly; keep extra broccoli for a fiber-rich side.

## Estimated Cost and Nutrition

- **Ingredient cost for 2:** about $5.50-$6.50 ($2.75-$3.25 per serving), excluding optional cheese;
  cost is prorated, not full-package checkout.
- **Per serving:** about 400-430 kcal, 39-42 g protein, 48-52 g carbohydrate, 8-10 g fat, and 10-13 g
  fiber. Optional cheese adds extra cost and varies by amount.
- **Basis:** assumes 4 oz dry chickpea pasta, 1/2 lb raw chicken, 1 cup marinara, spinach, and one
  small broccoli head. Nutrition is estimated from typical ingredient data and package labels; prices
  are rough U.S. online discount-grocery references checked 2026-10-12, not a local quote.
  [USDA FoodData Central](https://fdc.nal.usda.gov/)

These are all things I could do on my own, but I don’t usually have the extra bandwidth to spend the time thinking about it and writing it all out, much less putting it in detail in files and on calendars. So that’s what I had it focus on.

So Far, So Good

I’ve got to say, so far I’m very impressed. Every day it gives me an overview of my training load over the last couple of weeks, what my trend is, and how I should adjust my workouts. It’ll automatically update my workouts for me if it thinks they need to change. It knows the race I’m training for, and it knows my general workout plan. I use the 80/20 plan, and it looked up the details and integrated them into its strategy.

As for the recipes, I looked through them, and it’s not telling me to add glue or anything, so they pass the brief check. The one I’ve made so far was quite tasty and quite simple. It was pretty spot on for what I wanted.

Offloading Open-Ended Tasks

This is just an interesting example of the kind of task you can offload. Not just giving it something simple, but giving it an open task with some guidance and parameters, and then reaping the rewards.

All of this is running on Hermes, and for the most part I’ve got Hermes using GPT-6 Luna. It is very smart and very cheap.

More to come.