Last week I wrote about what we got wrong with AI in 2025. The patterns I saw companies repeat. The mistakes that stalled progress.
But that’s only half the story.
Because while most AI projects struggled, some succeeded. Not just in demos or prototypes, but in production. Delivering measurable value. Saving real time and money.
After working with growing companies throughout 2025, I saw clear patterns in what actually worked. Not theoretical use cases. Not vendor promises. Real implementations that changed how work gets done.
Here’s what I watched succeed.
1. Development teams shipping code 2x faster
This one wasn’t subtle.
Teams using tools like Cursor, GitHub Copilot, and Windsurf delivered faster than anyone expected. Not marginally faster. Dramatically faster.
DORA metrics showed it clearly.
Product backlogs that would normally take two or three quarters got done in one. Features that sat in the pipeline for months shipped in weeks.
The acceleration was real. Developers still wrote the logic, made the decisions, reviewed the code. But AI handled the repetitive work. The boilerplate. The syntax. The documentation.
Code quality didn’t drop. Velocity increased.
2. Processes running on autopilot
Marketing automation. Recuriting. Bookkeeping. Report generation. Data analysis. Monitoring. Anomaly detection.
These processes ran on autopilot with AI once teams knew exactly what they wanted.
The key phrase: once they knew exactly what they wanted.
When companies defined clear outcomes, set specific guardrails, and pointed AI at well-understood processes, it worked. It saved FTEs worth of time. It eliminated mundane work that would otherwise require dedicated team members.
For growing organizations with constrained resources, this was a lifesaver.
But it only worked when the target was clear. AI didn’t figure out what to measure or how to report it. Teams had to define that first. Then AI executed it reliably.
3. Design and prototyping accelerating product development
AI proved to be a surprisingly capable UI/UX resource.
Design teams fed it existing data, images, context, and personas. AI generated prototypes faster than manual design processes could. In some cases, it surfaced design patterns the team hadn’t considered.
The dependence on design researchers decreased. Not eliminated, but reduced significantly.
Teams still made the creative decisions. They still set the vision. But AI accelerated the prototyping phase in ways that compressed timelines and freed up designers to focus on higher-level thinking.
4. Notetaking, translation, and interpretation eliminating grunt work
Anyone who used AI notetaking in 2025 knows this: the grunt work of tracking meeting notes, action items, and follow-ups is gone.
But here’s what surprised me more: AI successfully translated conversations across languages. It interpreted intent, not just words.
I saw this work in settings where language barriers previously required human translators. Hospitals. Law firms. Global teams working across time zones.
Tools like Google Translate, Microsoft Translator, and iTranslate handled real-time conversation translation in ways that removed friction from routine communication. AI didn’t replace human translators for complex, high-stakes conversations. But for everyday exchanges, it worked. And it removed barriers that used to slow everything down.
5. Predictive analytics, monitoring, and IoT use cases continuing to mature
This isn’t new. Machine learning has been used for years in data prediction, monitoring, and analysis.
But in 2025, these use cases continued to deliver. Financial services. Manufacturing. IoT deployments.
These weren’t flashy. They weren’t making headlines. But they were reliable, proven, and generating real business value.
Where AI still hasn’t figured it out
Not everything worked.
Two areas in particular showed AI’s limitations clearly:
Customer support conversations fell short.
AI hasn’t proven itself in customer-facing support yet. Companies that put AI in front of customers quickly pivoted back to using AI as an assistant to the support agent, not as the front line.
Customers still want to talk to humans when they need help.
The technology improved, but the experience didn’t meet expectations. We crave human connection in moments of frustration or confusion.
AI-generated content and personalized outreach hit a wall.
Early in the year, AI-generated emails and outreach seemed promising. By the end of the year, the pendulum swung back.
Audiences can tell. AI-generated text feels similar. Generic. Impersonal.
Instead of engaging, it causes people to disengage. The human touch matters more than we thought.
The pattern in what worked
AI succeeded where:
- The problem was clearly defined
- The outcome was clearly known
- Guardrails were clearly set
- The AI tool was purpose-built for the use case
Not all AI tools are built the same. Generic platforms underperformed. Specialized tools designed for specific workflows outperformed expectations.
Development tools like GitHub Copilot and Cursor. Design tools like Midjourney and Figma AI. Meeting tools like Fathom and Fireflies. Marketing tools like Jasper and Copy.(ai)
Tools built for a specific job consistently outperformed general-purpose platforms trying to do everything.
The companies that succeeded didn’t throw AI at problems hoping it would work. They understood the work first, then deployed AI strategically.
Heading into 2026
AI isn’t hype. But it’s also not magic.
It works exceptionally well in narrow, well-defined use cases. It struggles in open-ended, human-centric contexts.
The advantage goes to companies that know the difference. That start with the work, define the outcome, and choose tools built for the job.
AI delivered real value in 2025. Just not everywhere. And not without discipline.
Most companies I see struggling with AI have the same foundational gaps. If you want to talk through where you’re stuck, let’s talk. I’d be glad to help.
I’m curious how 2025 played out for you. What did you try with AI? What worked? What didn’t?
If any of this resonates, I’d love to hear your perspective. You can reply to this newsletter or connect with me on LinkedIn and continue the conversation.
