Is AI actually making teams more productive?

Hello fellow humans - before we dive into today’s post, some updates:

For those that are new to our community, I’m Haroon - founder at AI Ready and the human behind this newsletter since 2017 (yes, way before ChatGPT).

I’ve spent more than a decade helping people put AI to work, picking up plenty of interesting experiences in that time, from:

Nowadays, my job is working inside companies to make AI drive meaningful business results.

Each week, I’ll share what I’m learning from this work and teach you how you can bring that into your own org.

Thanks for joining me on this journey.

Welcome (back) to Not A Bot.

IN THIS EDITION

Anthropic updated its Usage Policy for when AI makes decisions that seriously affect a person’s life (like hiring, loans, or medical advice). Those decisions now need a qualified person to check them, and people must be told AI was involved.

ANTHROPIC · OCT 8

Anthropic’s Haiku 5.5 costs about 75% less per task than Haiku 4.5. This is at a twentieth of Sonnet 5.5’s price per token. Test it on a repetitive task before you switch. Count review time (not just token cost).
ANTHROPIC · OCT 13

Among AI users, 68% of college grads say it makes them faster, compared with 51% of those with a high school education or less. And 52% of employees want more say in which tools get adopted.
GALLUP · OCT 6

I’m running a free, exclusive two-hour workshop for individuals responsible for AI adoption at their companies.

In two hours, I’ll teach you everything I know about AI at work - including what AI can do today and why it gets harder to use well across a team. Then we’ll map one recurring task from your own organization and decide on a first test.

We’re only accepting a limited number of attendees. If you lead AI at your company, be sure to apply.

You won’t want to miss this.

Last week, I met with two companies that were trying to figure out their next step with AI. Both were running into the same problem.

Their orgs were dividing into two groups: The AI-pilled people were experimenting constantly and changing how they worked. Everyone else was trying to keep up, sometimes unsure which tools their colleagues were using or where important information lived.

I hear versions of this often. As people get better at using AI, they build their own ways of working. They save useful instructions, collect reference documents, and sometimes build little apps to handle parts of their job. But their colleagues may have no idea how any of it works, where to find it, or whether they’re supposed to use it too.

That’s where the frustration starts for both groups.

The heavy users want everyone to adopt the approach that’s working for them. Others are still trying to understand what changed and how to participate.

Even when everyone uses the same AI tool, they can end up with very different ways of getting work done. Everyone can be putting in effort and still making life harder for each other.

1. People are developing separate ways of working.

People are making decisions independently about work that the team depends on. Which information should the AI use? What instructions should it follow? Where should the results go, and who checks them? Those decisions often stay inside someone’s personal setup until a colleague needs their work.

Imagine getting an AI-generated analysis without access to the sources or instructions behind it. You may have to redo part of the research before you’re comfortable using it. Your colleague saved time, but now you’re spending time figuring out what they did!

And more tools can make this harder to manage.

Each can have its own files, instructions, and permissions. Even a useful app someone builds can become another place where company information gets stuck.

I find it helpful to separate the four parts of collaborating with AI:

Teams need agreements about all four. Even if you’re using the same tool, it doesn’t settle the other three.

2. People have different levels of confidence and ability.

Some people hesitate because they don’t feel capable of keeping up. They may struggle to judge an AI answer or feel embarrassed asking questions their colleagues seem to have figured out months ago.

This gap affects everyone.

People who feel behind may avoid the new process, while the heavy users keep changing it without bringing their colleagues along.

Leadership needs to set the direction, and people need practical help following it. Both things must happen together.

Leaders should establish which tools the company supports, what information can go into them, and what needs review. They should follow those practices themselves. There also needs to be a clear way to try something new, especially when it stores company information or becomes part of work others depend on.

Give employees time to learn those practices using their own work. Training and office hours are useful here. So are internal hackathons, where people can see what colleagues are trying and get help building something themselves.

I’ve seen hands-on training help people get past that anxiety. When they practice on familiar work and have someone to turn to, they start seeing possibilities for themselves. They can also participate in conversations that previously felt out of reach.

This is why I’m extremely bullish on events for collaborative building, like hackathons.

They give people room to experiment together and expose problems you might otherwise miss. e.g., someone can’t access a document, two people are working from different sources, a useful set of instructions only makes sense to its creator.

Afterward, participants can turn what worked into something their team can keep using.

That means giving shared information and instructions a home, assigning people to maintain them, and agreeing where finished work goes. Customer records might stay in the CRM. Approved research might live in shared documents. Employees should know which sources to use without having to ask around.

Hackathons are a great way to “shock” the team into working better with AI - they are pound-for-pound one of the most effective tactics teams can use. (And that’s why we have a dedicated service to set this up for teams)

Often teams will want to jump into building an orchestration layer to standardize the complexity that results from uncoordinated AI usage across. By connecting AI tools into this layer, which can be used to maintain centralized resources like context and tools, teams are able to give their people more freedom to use different tools.

But it’s worth knowing that it can come with a heavy maintenance cost, as someone still needs to own this layer, the information it contains, and review updates.

tl;dr: An orchestration layer can be very helpful, but avoid making it more complex than it needs to be.

This week, talk to someone on your team who uses AI heavily and someone who’s having a harder time getting started.

Ask the heavy user: What have you figured out that would help someone else on the team? Could they use it today without your help?

Ask the person who’s struggling: What would make AI easier to use in your work? They might need practice, clearer rules, access to the right information, or simply someone they can ask without feeling embarrassed.

Once you gather their responses, identify one thing to fix. This could include things like:

You might be surprised by how much you learn about what your team needs from this simple exercise.

That’s it for this week’s issue of Not A Bot (formerly known as AI Ready, formerly known as Not A Bot). Thanks for reading.

If today’s story sounded relatable, apply to join the workshop I’m hosting next Friday.

Over two hours, we’ll take one recurring task from your organization, map where the work gets stuck, and decide on a practical first fix.

There are only 30 seats, and requests are open now.

Apply for your free seat

Until next time,