There’s a certain kind of AI planning session that feels like your watching Wile E. Coyote shop from the ACME catalog. The team has diagrams, arrows, five tabs open, and a very serious plan to automate an entire function in one heroic move. Then reality sprints by, beeps once, and leaves everyone staring at the whiteboard.
That’s the problem with a lot of AI advice right now. It jumps straight to full automation, as if every company should be trying to build a digital workforce by next quarter. Most teams don’t need that. They need three to five automations that are good enough to save time, cut repeat work, and help people make better decisions without turning the company into a science project.
That’s not small thinking. That’s product sense.
The phrase “good enough” sounds almost rude in tech because everybody wants the bigger story. Full automation. End to end workflows. Autonomous agents everywhere. It sounds bold in a deck. It also tends to collapse the minute it hits an actual Tuesday, when sales is still cleaning up CRM notes by hand, support is triaging the same tickets again, and PMs are still trying to turn six customer calls into one clear summary before lunch. A lot of AI teams make the same mistake. They go after the whole job instead of the most annoying part of the job.
That’s where the Wile E. Coyote analogy actually helps. He never looks for the simplest way to solve the problem in front of him. He builds a contraption. A huge one. Lots of springs, rockets, pulleys, smoke, and false confidence. Product teams do this too. They start with “How can we automate the entire workflow?” when the smarter question is “Where are people wasting time every single day?”
That one question saves a lot of money. Take the over-automated standup. Every team thinks it wants this for about five minutes. A bot collects updates before the meeting. Another tool summarizes them. A third one turns them into tickets. A fourth posts a tidy recap with blockers, sentiment, and action items. It looks efficient until the team still has to ask the same human question it always asks, which is, “Okay, but what’s actually stuck?” Good enough automation doesn’t try to replace the standup. It trims the dull parts around it. Summarize the call afterward. Flag the blocker that has shown up three days in a row. Link the relevant ticket automatically. That helps. Turning a live team ritual into a weird admin ritual for bots usually doesn’t.
Then there’s the prompt graveyard. Every AI-forward company has one. It starts as a shared doc full of useful prompts. Then it grows into a digital junk drawer. Some prompts worked once in a demo. Some only make sense if the original author is standing beside you. Some were built for an older model and now write like a confident intern who didn’t read the brief. And still nobody deletes them. This is where teams confuse collecting AI stuff with building AI workflows. A hundred prompts in Notion is not a system. It’s just a pile. The useful move is to find the few prompts people actually use under pressure, then turn those into real workflow shortcuts inside the tools they already live in. Put the sales email draft inside the CRM. Put the call summary inside the notes tool. Put the ticket triage prompt inside support.
If people have to go hunting for the magic, it’s not magic.
The same logic applies to the dashboard nobody reads. AI makes it very easy to generate scores, summaries, trends, and recommendations. So teams build beautiful dashboards full of movement and color and confidence. Then nobody changes behavior because of them. Leadership asks for a screenshot before a meeting, somebody crops a chart, and the dashboard returns to hibernation. That’s not intelligence. That’s decor. The real question isn’t whether AI can produce a better dashboard. It’s whether it can improve a real decision. Who should the rep call today? Which support tickets need human attention first? Which product issue is showing up often enough to deserve action this week, not someday? If the AI can help at that moment, inside the workflow, people care. If it lives on a separate tab waiting to be admired, they usually don’t.
That’s the big split. Demo value versus workflow value. Demo value gets the room excited. Workflow value gets used a month later when nobody’s clapping. The good AI product managers learn to care more about the second one. So how do you decide what belongs in the three to five workflows worth automating?
Look for frequency first. If a task happens all the time, even a modest improvement adds up fast. Then look for tolerance. If the AI can be imperfect and still be useful with a human checking the result, that’s usually a strong candidate. Then look at placement. If the output shows up where people already work, adoption has a shot. If it requires a new behavior, a new habit, and a new trust leap all at once, you’re making life harder, not easier.
This is less glamorous than saying you’re reinventing work. It is also how useful AI products get built.
There’s another part people don’t talk about enough. Every automation is a trust decision. You’re asking users to let software handle a piece of judgment, or at least the first draft of it. If it helps consistently, trust grows. If it fumbles in an expensive or embarrassing moment, people go right back to doing it by hand. That’s why guardrails matter. If an AI workflow touches customer communication, money, compliance, or anything irreversible, there should be a clear review path. If it can be wrong, users need an easy way to catch it and fix it. If nobody owns the quality of the workflow after launch, then the workflow isn’t a product feature yet. It’s a live experiment wearing business casual.
And yes, some things should stay human. Not because AI is bad, but because the cost of being wrong is too high, the workflow is too messy, or the judgment is too contextual. Plenty of companies are trying to use AI like paint over a cracked wall. If the underlying process is a mess, the AI usually just makes the mess faster and more polite.
That’s why “good enough” is a useful phrase. It forces restraint. It reminds you that the goal is not to automate everything eventually. The goal is to automate the parts that actually help.
I was reminded of this in a client meeting not long ago. They brought us in to talk about AI strategy, and within fifteen minutes someone asked the question every consultant hears now: “What can we automate, and how soon can we automate all of it?” That’s always the moment. You can feel the room leaning toward the big fantasy. Somewhere, an invisible ACME crate is being lowered from the ceiling. So instead of answering with a giant roadmap full of agentic dreams, we asked them to walk us through the work itself. Not the org chart version. The real version. Who touches what. Where time gets lost. Where people copy and paste. Where they wait. Where they retype notes from one system into another because two tools still don’t talk. Where a smart human is wasting an hour doing something a decent machine could do in three minutes.
By the end of that conversation, the answer wasn’t “automate everything.” It was much better. We found a handful of places where AI could actually pull its weight. Summarizing call notes into the CRM. Drafting follow-up emails that a human could review quickly. Flagging repeat support issues before they became a weekly fire. Pulling a short internal brief before account reviews so nobody had to play detective across six tools.
That was the advice. Don’t automate the entire company. Automate the recurring friction. The client looked a little disappointed for about ten seconds, right up until they realized this plan had something the giant plan didn’t. It could ship.
That’s the part people forget. A smaller automation that gets used is worth more than a grand system that lives forever in planning. The best AI product work usually doesn’t look like a moon landing. It looks like one less annoying task, then another, then another, until the team suddenly notices work feels lighter. That’s how you know you’re getting it right. Not when the architecture diagram looks impressive. Not when the dashboard glows. Not when someone says “end to end” three times in one meeting. When the work itself gets easier. That’s the field guide. Find the few workflows where AI can actually move the needle. Keep humans in the loop where judgment matters. Build for Tuesday, not the keynote. And if a plan starts to look like a rocket-powered bird trap with twelve moving parts, take a breath and ask whether you’re solving the problem or just drawing a better cartoon.
Share your favorite overbuilt AI story, especially the ones that looked brilliant until they met real users. Tag @cesarmorenoai if you post it, and subscribe at cesarmoreno.ai for more writing on AI, product sense, and how not to turn your company into an ACME test site.