From Android in 2009 to AI in 2026: two platform shifts

In June 2009 I started work as an Android developer at IndiaNIC Infotech in Ahmedabad. In 2026 I started Bolo Media, a company built on AI models. Seventeen years apart, the two moments feel oddly alike, and I think the mobile to AI platform shift has lessons for anyone building right now.

What Android looked like in India in 2009

It is easy to forget how young Android was then. The first Android phone had gone on sale in the US only in late 2008. Through 2009 Google shipped version after version, Cupcake, then Donut, then Eclair, each changing things developers had just learned.

In India, Android phones were rare and expensive. Nokia and its Symbian phones ruled the market, along with a huge base of basic feature phones. Most people were on 2G. Mobile data was slow and costly, and "app" was not yet an everyday word for most Indians. The Android Market was small, and a lot of what was in it was rough.

If you had told a shopkeeper in Ahmedabad that within a decade they would run their business through a phone, they would have laughed. Politely, but they would have laughed.

That was the platform I started on.

What a young platform feels like from the inside

A young platform has a particular texture, and anyone who has worked on one will recognise it.

The ground keeps moving. Every few months the platform changes. What you built last quarter may need rework, not because you did it badly but because the foundation shifted.

The tools are rough. Early Android tooling was famously slow and patchy. You worked around missing pieces and learned to accept that.

Fragmentation is constant. Different phones, different screen sizes, different versions. Testing was never really finished.

Most people around you are sceptical. The platform is new, so the obvious question is whether it will matter. Plenty of sensible people bet that it would not.

Nobody knows the right patterns yet. Best practice is being invented in real time, by everyone at once, mostly through mistakes.

I recognise every one of these in AI today.

The mobile to AI platform shift: what rhymes

Models change every few months. A prompt or pipeline that worked well in spring can behave differently after an update. That is the moving ground again. I wrote about living with it in building a product on AI models you don't control.

The tools are rough. Evaluation, monitoring and cost tracking for AI are all still being worked out. People are building their own tooling for problems that will feel solved in a few years.

Fragmentation is back, just in a new shape. Instead of screen sizes and OS versions, it is models, languages and output formats. A model that is strong in English can be weak in Gujarati. You test across that spread the way we once tested across handsets.

And the scepticism is back too, some of it fair. A lot of AI demos look better than the products behind them. Some of the doubt is earned.

What is different this time

The rhymes are real, but I do not think this is a rerun.

Distribution already exists. In 2009 a new platform had to wait for people to buy new phones. In 2026 almost everyone the AI product wants to reach already has a smartphone and a data plan. India's cheap-data years did that work. An AI product can reach a small business owner in a smaller town on day one, on the phone they already own.

The cost model is different. A mobile app, once built, cost very little to run per user. Many AI products pay real money every time they do something useful. That changes how you design, how you price, and how carefully you test before you scale.

Language matters more. Early apps could get away with English interfaces because the early users were mostly English-comfortable. AI that writes, speaks and creates has to work in the language the customer actually speaks. For Bolo that means Gujarati and Hindi, not as an afterthought. "Bolo" (બોલો) means "speak" for a reason.

The speed is higher. Android took years to go from curiosity to default. AI capabilities are moving faster than that, which makes the moving ground move faster too.

What I carry from the first shift

A few habits from the Android years still shape how I work.

Build for where the platform is going, but ship on where it is. It is tempting to wait for the tools to mature. You cannot. You ship on today's rough platform and design so the pieces can be swapped as it improves.

Keep the moving parts behind clean edges. On Android, people who isolated device-specific code had an easier time. With AI, keeping each model behind one interface means an update or a switch is a contained job, not a rewrite.

Test on the real thing. The emulator lied, in small ways. So do AI demos and benchmarks. The only result I trust is the one on a real input from the real use case.

Do not confuse early with wrong. Many people dismissed mobile because the 2009 version was clumsy. The clumsiness was temporary. The shift was not. I try to judge AI by its direction, not just by today's rough edges, while staying honest about which edges are actually rough.

The people a platform shift is really for

In 2009 the people who mattered most were not developers. They were the ordinary users who would one day put their whole working life on a phone. It took years, but it happened.

I think AI is heading the same way. The real change will not be in tech companies. It will be when a small shop owner can make a decent video ad in their own language, without hiring a studio. That is the bet behind Bolo Media.

I have written about the path between those two points, role by role, in from Android developer to founder. I do not know exactly how this shift will play out. I did not know in 2009 either. But having seen one platform go from curiosity to everyday, I have a fair sense of what the early, messy part feels like, and I am glad to be in it again.