- Able Craft & Code
- Posts
- The real story behind that viral MIT headline
The real story behind that viral MIT headline
95% of AI pilots fail—but not for the reason you think
Able Co
September 04, 2025
In this issue:
CRAFT: Why external partnerships are quietly winning the AI race
CODE: Why teams skip the fixes that actually work
NEWS: "Vibe coding" startups hit $100M ARR, Apple's AI desperation, and AI solving grassroots problems
CRAFT NOTES
What we’re thinking about
By now you've probably seen the headlines: MIT says 95% of AI pilots fail. Cue the collective groan from every CTO who's been pushing their team to embrace AI transformation.
But here's the thing—the headline misses the point entirely. "Almost everywhere we went, enterprises were trying to build their own tool,” said Aditya Challapally, MIT's lead researcher. On the flipside, the study shows that companies that purchased AI tools or partnered with third-parties succeeded about 67% of the time.
The reason is simple: It's hard to think outside the box when you're in the box.
“Generic tools like ChatGPT excel for individuals because of their flexibility, but they stall in enterprise use since they don't learn from or adapt to workflows,” Challapally noted.
That means you're essentially asking a race car to perform well on a horse trail—while ignoring that the trail itself needs to be rebuilt. Internal teams are naturally constrained by the pressure to deliver quick wins rather than address underlying data quality, integration challenges, and workflow redesign.
Success isn't just about outsourcing the problem. The right external partners don't just provide better technology. They force you to confront the unglamorous realities: Is your data actually clean enough for this use case? Do you have the infrastructure to support real-time integration? Are your teams prepared for the workflow changes this will require?
This isn't about AI being easy—it never was. It's about being smart with how you approach the hard work. The 5% of companies that are succeeding understand something the other 95% missed: AI transformation isn't a technology project that you can shortcut with the right tools. It's a foundational capability build that requires both strategic thinking and tactical execution.
While 95% of companies are still trying to make AI fit their existing processes, the smart money is on redesigning the processes first.
 |  |
CODE LAB
What we’re discovering
"When we work on performance optimizations, it's tempting to go for the biggest and most complex fixes first, assuming they'll have the greatest impact. However, I've discovered that the simplest changes—like fixing N+1 queries, switching to collection rendering in Rails, or simplifying cache keys—often deliver the most noticeable improvements. The lesson for me has been clear: start with the basics and let the product itself show you when something more sophisticated is truly needed ."
Why this matters for your team: Before building AI-powered monitoring systems or complex caching strategies, check the basics first—are you asking your database the same question repeatedly? Are you loading more data than you need? These fundamental improvements typically provide immediate, measurable gains and create a solid foundation before adding advanced solutions. The principle applies beyond code: in AI adoption, start with clean data and clear processes before implementing sophisticated tools.
AI NEWS
What we’re paying attention to
While "vibe coding" works for startups building from scratch, it's a different challenge than enterprise transformation. Swedish startup Lovable hit $100 million in annual recurring revenue just eight months after launch using AI that generates code from natural language descriptions. Competitor Anysphere reached a $9.9 billion valuation with similar tools. The success highlights how AI can accelerate greenfield development, but enterprise transformation faces fundamentally different challenges—integrating AI into existing workflows and organizational processes that coding tools alone can't solve.
Analyst Dan Ives bluntly assessed Apple's AI position: "They're playing baseball without a short stop and a catcher. They're missing AI" and suggested the company will need to acquire rather than innovate its way out of the problem. Ives specifically mentioned Perplexity as a potential target ahead of the upcoming iPhone event. The assessment underscores how even tech giants with massive R&D budgets are finding it difficult to keep pace with AI evolution, potentially forcing historically build-everything-in-house companies to fundamentally change their acquisition strategies.
While Silicon Valley debates coding tools, AI is tackling infrastructure planning in Rwanda, providing agricultural advice in 27 languages to over 110,000 African farmers, and mapping 75 million miles of unmapped waterways. Ghana-based Darli AI delivers farming guidance through WhatsApp in local languages, while AI tools identify ideal bridge sites to connect rural communities. These applications demonstrate AI's potential beyond enterprise efficiency—directly addressing basic infrastructure and access challenges that have persisted for generations.
SKIP THE ENDLESS EXPLORATION, START BUILDING
Done with endless AI meetings? Ready for practical implementation?