In this post Alex layed out a nice framework for thinking about how to get started on some proven AI use cases for your business. These "getting started" cases are great because they are easy to implement and can often yield quick wins for people it various roles. But the marginal improvement for most of these cases is not dramatic - maybe finding some piece of info goes from 10 minutes to 2 minutes. Or drafting that sales email takes 5 minutes instead of 15. In most cases these aren't adding up to "hours and hours" per week.

Following are some ideas for going beyond the basics. How can AI solve actual "jobs to be done" and start saving my staff significant time?

Empowering individuals

I was talking with a friend who is building an "AI sales coach", and I offered my assumption that his product must be mostly helpful for junior salespeople. He replied that in fact many of their most committed users were extremely experienced salespeople.

The truth is that AI in its current form shows up as tools, and tools reward people who put in effort to learn how to use them. In fact I consider most AI to be essentially "power tools for knowledge workers" - they help people write, think, and build faster than they could without the tools. (There are certain jobs that boil down to "essentially language processing" - like being a language translator - and those jobs are in fact being automated by AI. But those are really niche cases.)

So one way to get more from AI is to encourage and unleash the power users on your team who can show others how to use the tools more effectively.

But to get more concrete, here are some specific ways that I've observed individuals moving faster with AI:

Job: writing and testing code This one is super obvious, and software devs have been early adopters of AI tech. But many folks are "just" using Copilot - getting nice green field code suggestions or inline completions. But there are much higher leverage ways to use some tools out there. Our team really likes Claude, and we've find some great use cases:

• Design co-pilot. Interactive designing of new components (like a Chrome extension) by starting with a nice design spec, having the LLM generate the code according to the spec, and iteratively reviewing and tuning the implementation using plain English.
• Explain this code. Claude can read and explain code that is quite complex. Claude can analyze and document code in large enough chunks to very effectively document software architecture or complex parts of your software system. It can even generate Mermaid diagrams to accompany text descriptions. We've used this approach to document more complex parts of our system and then added the documentation to the project.

Job: answering questions at work Lots of companies have invested now in "Enterprise Search" AI systems which are based around hoovering up as much data in the company as you can and packing it all into a single expert chatbot. Usually the target is to help the Customer Success team answer questions (or expose the chatbot directly to customers).

But more and more I see individuals and particular teams building their own "personal knowledge valet" trained from their documents, emails, and Slack messages, and connected to their systems live. This personal use case can be quite powerful because limiting the scope of info fed into your Knowledge Bot helps it produce answers that are much more targeted and accurate.

Job: executive assistant At most companies only the most senior execs can hope to employ an "Executive Assistant" who helps manage scheduling, gathers information and reporting, and logs and summarizes meetings. But AI is introducing the ability now for everyone at a company to have their own EA. Your AI can attend and transcribe meetings, record ad-hoc notes from emails or online discussions, or help you craft project plans.

Empowering teams

Job: analyze data to improve decisions Well functioning teams seek to ground their decision making in quantitative data. In larger orgs this usually means the "BI Team" and "Data Engineering Team" are devoted to sourcing, cleaning, curating and reporting on analytics data within the company.

However, smaller companies usually can't afford a dedicated BI Team. And even in those larger orgs the need for the right data, right now, often outstretches what the BI Team can deliver. The result is PMs, EMs and individuals spending time with various ad-hoc tools trying to extract the info they need to make good decisions.

This is a burgeoning area for AI tooling, where AI systems can be connected to live systems and data warehouses and provide support for ad-hoc analytics, including generating SQL on demand from natural language.

Job: data processing tasks Many companies still require people to spend a lot of time processing data. The accounting team is dealing with invoices and receipts. HR usually has lots of documents to manage benefits and employment compliance. The legal team is all contracts, all the time, working with the security team to fill out security questionaires and compliance docs. Many of these tasks can be made easier/faster by using AI to help with document and language processing.

Getting to Task automation. One of the big goals of newer AI agent systems is to combine document+data processing with automation. Agent systems can apply not just document processing and language understanding, but conditional logic and integration with other systems. This holds the promise to help automate "repeatable knowledge tasks" that can often consume lots of team time. At the moment these solutions are often "best effort" due to LLM hallucinatons and other limitations, so they are best applied to tasks that don't require 100% correctness. Or the automation has be designed so that people can oversee it and correct any mistakes.

Things like filling out product descriptions for an Ecommerce catalog, or doing research on new customer leads, or competitor analysis for a marketing project. These kinds of tasks can be partly solved by the AI and bolstered by people effort.

Job: improve team performance Using AI for coaching sales people is currently in vogue, but I expect "AI performance coaches" to proliferate to other teams as people get more comfortable with AI coaching and sensitivity and contextual awareness of these systems improves.

Custom applications

Gen AI technology can be useful across many different applications, and some of the most impactful uses will be very specific to a particular company. Adopting a platform that includes lots of capabilities (multi-model support, connectors and tools, extensible RAG) and flexible integration options is a great way to enable discovery and implementation of custom use cases without resorting to expensive custom development.

‍