Plant a Thousand Seeds: How Teams Find the AI Workflows That Stick

At Webflow Conf in Boston, one topic kept coming up in conversations and sessions our team listened to.
Stop trying to pick the perfect AI tool for the whole company from the top.
Give people room to experiment, see what grows, and scale what works.

It sounds simple. But it answers the question a lot of companies are stuck on: everyone has access to AI, so why isn't the organization as a whole getting better?

Part of the answer is that the experimenting is already happening. MIT NANDA's State of AI in Business 2025 found that while only 40% of companies had purchased an official LLM subscription, workers from over 90% of surveyed companies reported regular use of personal AI tools for work. BCG's AI at Work 2025 survey found that when employees don't have the AI tools they need, more than half will find alternatives and use them anyway.

So people are testing. They're just doing it alone, in personal accounts, where nobody else learns from it.

Wharton professor Ethan Mollick explains why that matters in his piece on Leadership, Lab, and Crowd. AI that boosts individual performance doesn't automatically improve organizational performance. Getting there requires rethinking incentives and processes, and nobody has a ready playbook for how AI should work inside your specific company.

The people who know your workflows best are the ones running them. That's why the seed approach makes sense, and the way it is described follows four steps.

It starts with planting the seeds. Teams get access to tools and clear permission to try them on real work. Mollick recommends defining areas where experimentation of any kind is allowed and leaning toward permitting AI use wherever it's ethically and legally possible. Leadership backing makes a measurable difference here: in BCG's data, the share of employees who feel positive about GenAI rises from 15% to 55% with strong leadership support.

Then you let them grow. People test tools on their own tasks and keep track of what helps and what doesn't. Experienced employees are well placed for this, since they can judge through trial and error when AI is useful for their work in ways outsiders can't.

Next comes sharing the harvest. Teams meet in regular work sessions, show their results (failures included), and narrow the list. These sessions only work if sharing feels safe. Mollick notes many workers hide their AI use, worried they'll lose respect, get no reward, or see productivity gains turn into layoffs or heavier workloads.

Finally, you plant what works everywhere. The shortlist becomes a shared standard, documented and built into how the team operates. MIT's report describes forward-thinking organizations learning from shadow usage and analyzing which personal tools deliver value before procuring enterprise alternatives. BCG found the real value comes when companies reshape workflows end to end, since adding AI tools to existing ways of working isn't enough.

If you want to start this month, the first session doesn't need much. Get your team in a room and ask four questions:
1. Which AI tool or prompt saved you real time in the last two weeks?
2. Where did AI slow you down or make the output worse?
3. Would this work for someone else on the team, or only for you?
4. What would need to change in our process for everyone to use it? Anything that passes question three goes on the shortlist.
Anything that passes question four becomes a workflow. Then repeat every few weeks, because the tools keep changing.

Where this gets harder is engineering. A marketer can adopt a new prompt by Monday. A dev team adopting an AI workflow touches repositories, code review, CI pipelines, and security rules. One engineer's great AI coding setup becomes a team risk if it spreads without shared conventions and guardrails.

Our take is that organic experimentation is the right way to discover what works. Turning "this works for me" into "this is how we ship" needs engineering structure: agreed configurations, review standards for AI-generated code, and clear limits on what agents can touch. That step is where most of the long-term value sits.

Your team is probably already growing a few AI workflows worth keeping. Which ones are ready to become standard, and what's holding them back?