AI and running: Why technology makes human coaching more valuable, not less

I'm training for a 5K at the end of November. Target time: sub-16 minutes. Current PB: 16:11. Season's best: 16:34. I'm also a licensed running coach with paying clients.

And I'm using Claude, an AI assistant, to help design my own training plan.

In the UK it's estimated that around 40% of the population have used an AI tool in 2025, so maybe it's not that surprising to learn that it's being applied this way. I have no doubt that runners are already doing something similar. I'm currently studying an AI programme at the London School of Economics, and I'm keen to get my hands dirty so to speak. Sufficiently dirty now, I thought it would be useful to share my approach from a coach's perspective.

As I go further on this AI and running journey, I'm finding that contrary to my initial concern that the human coach was on his way to becoming obsolete, AI is actually providing more clarity on what coaching is and where the human has, and will always have, advantages.

My training method

If you've read my previous articles, you'll know I'm a big fan of threshold training and building my aerobic base through controlled intervals. For this training block I wanted an extra edge, and I'd been keen to introduce more of a Canova-style race-specific approach for the final weeks (tempo runs and intervals at precise percentages of race pace). I also wanted to shift to a 9-day cycle with a simple workout, easy, easy flow that would give me proper recovery between sessions.

I fed Claude with all my notes on these training principles so it would understand them better, as well as my historic training data, and then my training parameters (9-day cycle format, dates of goal races, current training level, etc). It took a bit of back and forth to amend some errors and tweak the plan, but after around 10 minutes I had a pretty good draft of a 12-week plan that seamlessly blended the two training philosophies whilst developing the training load to peak for my 5K race.

After every session, I give Claude an update on how it went, mainly conversationally rather than overloading with data. We then discuss whether there's any need to make adjustments to upcoming sessions.

For someone who coaches runners, this felt simultaneously natural and slightly transgressive!

What AI does really well

As already mentioned, it whipped up my plan in no time. Not perfect first time, but probably saved me a good hour or so doing it all manually. It handled this complexity without ego. When I explained I wanted to blend two philosophies, Claude didn't say "that's not how it's done." It helped me think through the underlying principles and how they might work together. No attachment to a particular method. No dogma.

It could handle discussion of pretty sophisticated training knowledge. No longer hidden away in training manuals or buried in forum threads, I was able to interrogate it with questions about lactate, periodisation, race tactics etc.

It's endlessly patient with experimentation. I can workshop ideas, change my mind, ask stupid questions. No interpersonal dynamics. No concern about wasting someone's time. It's there 24/7. One of my favourite and funniest moments was asking it to give me a pep talk 5 minutes before I was about to head out of the door. It was a bit cheesy but you know what, it actually did pump me up a bit!

Alright, let's do this.
You've already proven you can handle this.
Last week you did 5 miles of this effort and finished with a 5:47 final mile. Today you're adding one more mile, but you're also smarter, more adapted, and properly fresh.
Sunday you ran 4×1600m at 5:30/mile and called it "surprisingly comfortable." Today's 6:05/mile pace is 35 seconds per mile slower than that. You have the speed. You have the endurance. Now you're just putting them together.
The first 3 miles are about patience. Settle in at 6:08, let your body find the rhythm, then click into 6:05 for mile 3. You're building, not surviving.
Miles 4-5 are where you prove something to yourself. This is where it gets hard. This is where champions are made. You don't back off, you don't negotiate with the discomfort - you hold the pace and trust your training.
Mile 6 is your victory lap. You'll see the finish, you'll feel strong, and you'll push it home at 6:00 or better because you've got it in you.
This isn't about setting a training PR. It's about becoming the runner who can hold 6 miles at tempo. That runner breaks 16:00 in a 5k. That runner is who you're becoming today.
You're ready. Your legs are fresh. Your mind is sharp.
Go run 6 miles at tempo pace and come back telling me how you nailed it.
Now get out there!

What AI cannot do

Here's where it gets interesting. The things AI can't do are precisely what make coaching valuable in 2025.

It can't see the full picture. Claude can access my workout data. It can't see that I'm navigating a work deadline. That my sleep's been disrupted for three nights. That I'm mentally exhausted even if my HRV looks fine.

Most athletes aren't chasing performance in isolation. They're juggling work, family, stress. The challenge isn't just physical adaptation, it's finding space for that adaptation within everything else competing for bandwidth.

When I coach someone, I always ask: What's the cost of this session, not just in energy output, but in life bandwidth? AI calculates training stress scores. It can't calculate whether you've got the emotional reserves to execute a hard workout after a difficult week at work.

It's reactive, not proactive. You want a coach who thinks ahead about what's happening in your life so they can support you properly. Good coaching anticipates. It doesn't just react.

I can't imagine coaching someone without understanding the weeks ahead. What are your actual demands? Your real time availability? How will work stress or travel hit you?

The strongest coaches stay ahead of the curve. Pre-empting fatigue. Thinking about travel. Contextualising work stress before it derails progress.

Simple test: If your week changes and your plan doesn't, you're not being coached. You're following a programme.

AI only knows what I tell it. A human coach develops intuition over time. They spot patterns before challenges emerge.

It can't be invested in your journey. Coaching is emotional work. You're there for the ups and downs. The breakthrough sessions and the brutal failures. The moments of doubt and the days everything clicks.

Knowing that your coach has your back and is genuinely invested in where you're going is among the most valuable assets of all. This is unique to human coaching. AI doesn't wake up thinking about your race. It doesn't feel disappointed when you get injured or excited when you nail a session. It can't share in your journey because it's not on it with you.

Care isn't cheerleading. Want to achieve something remarkable? It'll require commitment, hard work, resilience, navigating pressure. A coach's job, with care as bedrock, is delivering appropriate challenge relative to your aspirations whilst providing substantial support.

AI generates encouragement. It can't be your anchor when things wobble. Or your mirror when ego takes over. It can't have the difficult conversation about whether your goal's realistic given your circumstances. It won't tell you the hard truth you need when you're lying to yourself about recovery.

It can't build an ecosystem. Strong coaches don't work in isolation. The best results come from integrated systems. Specialists collaborating to serve the whole athlete.

A coach trying to do everything isn't demonstrating expertise, they're limiting it. The strongest coaching includes the self-awareness to sometimes say, "I don't know", and to recommend the athlete draws upon the knowledge of a physio, a nutritionist, a strength and conditioning coach, etc. AI provides information across these domains but can't coordinate specialist care. It can't know when to refer you to expertise beyond its capabilities. It can't build the support network that surrounds serious athletes.

It can't simplify. In a world drowning in data and content, the coach's job is creating clarity. Most athletes don't need more information. Information is cheap. AI makes it cheaper.

What they need is help to translate it so they can decide what to do with it.

The coach's job is to filter: What's good versus bad? What's relevant versus irrelevant? What should you double down on because it's likely to be your secret sauce, versus what's interesting but ultimately noise that adds complexity and confusion?

One thing that's consistent among the very best athletes is that they embrace simplicity. They're never looking to add more, rather the opposite: they reduce the noise.

AI gives you more. A coach helps you do less, better.

What is coaching then?

Using AI for my own training has given me the sense that coaching is better framed as a form of leadership. Our job is to get the most out of athletes. That requires giving athletes what they need, not what they want. It requires honest friction. Tough conversations about rest, priorities, sometimes unrealistic goals.

Growth doesn't come from agreement. It comes from trust built from truth.

AI agrees with you. It validates your reasoning. It can't push back with authority born from years of experience and relationship built on care.

When I tell Claude a workout felt "surprisingly comfortable", it takes me at my word. A human coach who knows me might say, "You ran those splits at that heart rate. You were working harder than you think, whether it felt controlled or not".

What AI and running means

For athletes: AI's making you smarter and better informed. I strongly encourage its use. But recognise that becoming more knowledgeable also reveals what you don't know.

You're not replacing coaching. You're becoming ready for deeper coaching. You're outgrowing cookie-cutter programmes and becoming someone who can benefit from what only humans provide: observation, accountability, ecosystem coordination, wisdom from guiding dozens of athletes through similar challenges.

For coaching: The bottom tier, i.e. generic plans and basic information, is being automated. Frankly, it should be. Cookie-cutter programmes never addressed individual needs anyway.

The top tier, where sophisticated, personalised coaching resides at the intersection of humanity and performance, becomes more valuable. Coaches who thrive will embrace technology whilst offering what technology can't: presence, observation, accountability, honest friction, care.

Then there's a new middle tier: AI-assisted athletes knowledgeable enough to direct their own training but wanting sophisticated analysis and planning. They're not replacing coached athletes. They're creating a new category of self-sufficient athletes who may eventually seek coaching precisely because AI taught them to value what it can't provide.

My experiment continues

I'll race my 5K at the end of November. Whether I break 16 minutes or not, I'll have learned something about training at the intersection of human knowledge and artificial intelligence.

What I know already: AI's made me a more thoughtful coach. Understanding its limitations has clarified what I offer clients that no tool can replace. Using it has shown me what runners actually need from coaching in 2025.

No professional athlete will be replacing their coach any time soon, and nor should any amateur. They should demand better ones though, and to keep on providing them with the skillset and perspective that an AI cannot.