Ankit Anand

I write about SaaS and the customer side of business | Worked before in fintech, adtech, healthtech | Learned a few things @UW @IIMBangalore @NITTrichy

[Ankit Anand @a16anand]

@ananis25 👀

[Garry Tan @garrytan]

This is the simplest distillation of what I have learned about agentic engineering this year

Push smart fuzzy operations humans do into markdown skills. Fat skills.

Push must-be-perfect deterministic operations into code. Fat code.

The harness? Keep it thin.

[Ankit Anand @a16anand]

Muse Spark's capabilities align with Meta's social media roots: create image, video or simple games like tic tac toe, monopoly, etc. The output is neat and produces quite quickly. Much better than Sora. Given Meta's growth mktg prowess, wud be keen to see how quick it grows.

[Ankit Anand @a16anand]

@heyrimsha @garrytan I am sure ppl realize this but pre-LLMs, I had to spend the initial 3-4 interviews validating market hypothesis. Using AI as a thinking partner does short-circuits that process.

[Ankit Anand @a16anand]

@heyrimsha @garrytan The devil's advocate question is super helpful. I would still argue founders should spend actual hours talking to customers. Discovery interviews lead to uncovering insights not documented anywhere, and a likelihood of converting some to testers and eventually paying customers.

[Ankit Anand @a16anand]

@aakashgupta Thanks for sharing. Def worth including in the CS stack at scale.

[Ankit Anand @a16anand]

@Brocklesnitch Funny enough and I am not saying doomscrolling is good, but even social media content starts becoming boring in minutes.

[Ankit Anand @a16anand]

Good short read even if you don't care about investing per se. Particularly the principles of success - even if you are not the most talented one or have the best ideas, you can still create an edge by outworking others and following repeatable processes. https://t.co/QNXbzLWy3M

[Ankit Anand @a16anand]

Most vertical AI players currently use a bunch of cloud providers. While switching to just GCP doesn't make sense in the short term until Gemini models (w/ TPU architecture) get priced significantly lower, I would be curious how Gemini's performance affects this over long term.

[Ankit Anand @a16anand]

Although Google has significant lead over Amazon and Microsoft in the race to build in-house Application specific chips with TPUs, it still faces significant headwinds from Nvidia due to the latter's ecosystem driven network effects. https://t.co/KpNTicEV7C

[Ankit Anand @a16anand]

The above article was insightful as it uncovered both aspects - how TPUs can be cheaper and energy efficient for vertical applications/inference focused, and how using it may require an overhaul of the code base (also a potential risk of lock-in on Google's cloud).

[Ankit Anand @a16anand]

Google's latest model Nano Banana Pro is widely talked about. So I tried creating a marketing banner on it. With clear instructions on copy (header, body, CTA content), the output was professional grade but the resolution was sub-optimal. Clearly not usable in an actual campaign.

[Ankit Anand @a16anand]

If the model performs better, and it should given its context length of almost twice of the GPT 4 model, it will increase the adoption significantly and propel ChatGPT to Bn+ WAU within the next 2 months. It's quite costly bet though. 👀

[Ankit Anand @a16anand]

10 yes back Flipkart deprecated its website to drive users to its app. As its competitors gained mkt share, it rolled back within 5 mos.

Open AI's move to deprecate previous models in favor of GPT-5 reminds me of the same. It's a big and expensive bet. 🧵

[Ankit Anand @a16anand]

Hope there is no band playing at the conference. 🤣

[Astronomer @astronomerio]

Thank you for your interest in Astronomer.

[Ankit Anand @a16anand]

@gammaapp @OpenAI Wonder if every customer side AI interface tightly holds context, is there a scope for a portable context wallet? Identity+context that users can carry around. Account Aggregator framework in India has such model where users own their data & can select which entity to work with.

[Ankit Anand @a16anand]

I used to love making slide decks before as it helped me frame ideas in a succinct and easy to explain manner. And it felt like a defensible skill. The thing that I wasn't great at was design - fonts and images to use. 🧵

[Ankit Anand @a16anand]

@gammaapp While I understand the logic behind lack of interoperability bw different LLM interfaces (context is the new moat for customer-facing AI tools), moving within the same tool shouldn't be an issue. Or maybe there is a technical reason it doesn't work? @OpenAI