Do's and Don'ts of AI Implementation (From an Expert)
I got to pick the brain of an AI expert.
And no, it wasn’t Claude. 😂
It was Rebby John, the incredible founder and CEO of Fastlane AI (spoiler alert: Atlanta Ventures Studio company!), who helps mid-size companies understand where and how to get started with AI.
What can startups learn from mid-size companies using AI?
A lot actually!
Here are the top tips from Rebby on what to do (and not do!) to make the most of AI at your company!
4 Dos For Implementing AI
1. Do identify where you are.
Two types of companies:
Know what they want. Need manpower or expertise to execute.
Know they “need AI.” Don’t know what’s possible or where they should start.
Not sure they want or need AI.Literally does not exist anymore. Use or perish! (Note: startups have known this for a while 😉)
Which category do you fall into?
It’s good to get clarity (and be honest with yourself).
Different plan of action for each!
2. Do put your best people on it.
How do you succeed with AI?
Put your best people on it.
Even when they have other things that are really, really important, the successful companies are taking their best people and reallocating them to AI projects.
At a large company, your best people are your busiest.
At a small company, everyone is the best and everyone is busy. 😂
BUT…the idea is that companies using AI successfully are making intentional tradeoffs.
Have no time?
Make time. THIS is the important project.
In Startupland, it means deprioritizing very important things for AI.
Painful in the short term but worth it 10x over long term!
3. Do expect multiple iterations.
“AI is not good and really annoying” - me, in 2022, after one try with a terrible prompt.
Reminder to all companies — don’t be like me! 😁
Of course it wasn’t good. I expected perfection on the first go round.
The best companies understand that the first implementation is not the final outcome!
One great thing about AI is how quickly you can refine and tweak.
Companies willing and planning to iterate will have the most success!
Startups already know this deeply. 😂
The more you expect to suck the first time the faster you can get better!
4. Do rethink the process.
2X BETTER: Take current process. Add AI.
100X BETTER: Change the whole workflow!
Companies see most value when they’re willing to reimagine what’s possible.
If you were starting from scratch, what would it look like?
If you had no employees, what would you do?
If you could only use AI, what would be different?
Broaden your perspective to get the most out of AI!
4 Don’ts For Implementing AI
1. Don’t expect AI to do everything.
Love this action step from Rebby:
Make a list of pain points, manual tasks, any areas AI might be able to help with.
Then get honest about what AI can and can’t do.
Put your list in an AI tool or ask an AI expert for feedback!
AI is great, wonderful, amazing, but it can’t do everything and some things are going to be faster, easier wins than others.
Better to know that up front so you don’t force it! (Or drive yourself crazy 🤪)
2. Don’t make it harder.
Yes, custom agents are cool. But if there’s a SaaS tool that can already do it, no need for agents.
Opt for simplicity!
I love this decision flow from Rebby and Fastlane AI:
tl;dr: use SaaS → SaaS can’t do it? Use no-code/low-code automation → Low-code can’t do it? Build custom agents.
3. Don’t judge based on technical ability.
One of my favorite stories about AI implementation from Rebby was a woman who was not technical, held the same job for 30 years, and she figured out how to automate 75% of her daily tasks using AI!!!!!!!! 🤯
Did she work herself out of a job?
Kind of.
She got promoted!
Fastlane showed her how to use Claude to write Python scripts and she was off to the races.
What used to take 2 hours took 2 minutes.
That’s the ah-ha that makes people believers.
It’s not only engineers or tech-saavy folks.
The most important qualities for AI success?
Willingness to try and openness to learn and iterate!
4. Don’t decentralize AI agent creation.
Give everyone the keys to the AI castle! Build whatever agents you need!
In theory, yes.
In real life, it leads to lots of half-baked agents that aren’t helpful.
Yes, you want everyone to have access to approved and connected Gen AI tools (e.g. Claude or Gemini team account synced with your Google Workspace).
Yes, you want to decentralize identifying agentic AI use cases.
But companies get more value when the responsibility for building AI agents is centralized.
At a mid-size company, it’s a Center of AI Excellence, a hand-picked team that builds, verifies, and rolls out AI agents (and other AI initiatives) to the company.
At a startup of 10+ people, designate a “directly responsible individual” to create agents so your team’s AI implementation is working and aligned.
Yes, everyone is using AI and you want people to be empowered and sharing ideas.
But from the front lines of AI implementations, Rebby sees a truth we all know well:
Ownership and accountability result in higher quality output!
What has worked for your company?? What’s your best AI tip? Do you have a favorite Gen AI tool or use them all?!?
The centralized agent building point is such a good one, and it's been our experience at Larken too, even with a much smaller team. I've taken ownership of the build because we need consistency and quality, rather than a bunch of half-baked disconnected agents, but we do a lot of work before and during the build collaborating. I spend a lot of time with each team member understanding what's actually slowing them down, where the friction lives in their day to day, and what AI could realistically take off their plate so they can spend more of their time on the parts of their work that genuinely require their judgment and experience.
Once we know what to automate and why, building the agent and wiring up the data sources is the easier piece and we get much better outputs. I'd be curious to hear whether Rebby sees the same dynamic playing out in mid-size companies, or whether the calculus shifts at a different scale!