Information workers can no longer build their career on asymmetric process knowledge. The person in the office that knew how to work the niche corporate workflow to get their way will be replaced by documentation and an MCP server.

A large overhead for new employees is understanding processes at a company or relying on someone to inform them how things are done. The archetype of that odd senior taking pride in the ability to recall the weird edge cases of the internal admin portal hasn't left yet.

Products will change

Products that avoid allowing their user to voice demand, intentions and needs; will soon feel like exploring a webpage built in the 90s, cute, quirky and stuck in the past.

The need for an experienced power user has been replaced by a Claude or a Gemini. If the LLM can't do it, should we deprioritise it and focus on the things it can do? We've never had time to explore product features and it doesn't help when the UI isn't sorting the buttons by their relevancy to the problem I am having, there's a gap here and your competitor might be offering me a quicker route to market.

UX design and effort is built around imitating interfaces that are already familiar to the audience, for example we still use keyboards optimised to avoid jamming the hammers in a typewriter, a problem modern keyboards no longer have. There's a learning curve that needs to be overcome when using systems, new and old.

The interface can create friction.

We are shifting our interactions with the product to be ask and you shall receive.

If you aren't empowering the ordinary to become power users over night you continue to have an under utilised product with features that customers need but can't find. Search is more relevant than ever.

Don't leave features on the table. Create the power customers.

As an aside my grandmother no longer needs me to teach her how to upload images to instagram. It would appear her ChatGPT assistant has taken my job.

Learning will change

Software has had the most prevalent increase in power users; here the iteration cycle for learning has diminished significantly. If you've built your career around building products through proof, developed learnings and iteration, that muscle now allows you to run micro tests 500 times a day.

The gap self learners and builders are creating will only increase as we've exposed them to a tutor with 24/7 office hours. Some might argue that the models are not deterministic but I'd lean more towards a computer that has read the entire corpus of human experience for references than a tutor that hit their Goodreads target of 12 books this year.

Anecdotally; the cycle time of learning to code used to be about 20mins of searching or syntax look ups. That has been cut down to seconds and if OpenAI is giving us garbage we can test if it works quicker than it can write. There are few industries that can give feedback on a decision as quick. In some industries you have retired by the time a decision unravels.

The speed at which adopters can upskill has plummeted but that's only if they're conducting a piecemeal development approach and not generating an entire slop product or forgetting why there are files in a folder called src/.

As with science there's no way to determine what change led to what outcome if we are changing everything all at once. Don't replace ceteris paribus with prompt-and-pray.