Kiran Kulkarni

Our stories about AI

By Kiran Kulkarni on July 26, 2026

In my world there are two stories we tell about AI.

Either it’s coming for your company, or it’s going to triple your revenue.

Threat or multiplier. Fear or greed.

Both stories are about the organization. Neither is about the person.

Last week I sat in a room with lawyers and policymakers at Oxford, and they were telling a third story. Not “what will AI do to us?” but “what do we owe each other when we build it?”

That’s a humanist question. And it turns out it’s also the security question.

Because as I mentioned to them, securing AI isn’t one job. It’s six:

– Data — protecting privacy and sanitizing what the model learns from
– Model — fine-tuning, protection, and adversarial testing
– AI Application — validating inputs and outputs, controlling agents and plugins
– AI Infrastructure — hardened cloud, identity, and access
– Assurance — monitoring, red-teaming, hunting vulnerabilities
– Governance — policy, risk, and knowing what AI you actually own

Engineers start in the middle. Policymakers start at the end.
The gap between them is where trust leaks out.

https://www.linkedin.com/company/oxford-media-policy-summer-institute/posts/?feedView=all

Can AI Replicate the Nike Standard in Ads?

By Kiran Kulkarni on April 6, 2026

In the world of advertising, Nike is the undisputed gold standard. Their campaigns are high-budget cinematic experiences born from the “Studio Production Model”—massive sets, world-class crews, and months of post-production.

For years, this level of quality was a “walled garden,” accessible only to brands with million-dollar retainers.

My partners and I recently took a classic Nike campaign and attempted to recreate its essence using a professional Gen Media workflow. Below is the original, followed by my AI-augmented recreation.

ORIGINAL NIKE AD

The Gold Standard: Traditional Studio Production.

RECREATED VIA AI

The New Standard: Expert-Led Gen Media Production.

At first glance, you see a high-fidelity ad. I see an AI workflow that solves the three biggest problems facing marketing departments today:

1. The Death of “AI Slop”

Most people think AI generation is as simple as typing “Nike-style ad” into a prompt box. The result is usually “AI Slop”—generic lighting, and a cheap “plastic” look that ruins brand trust.

Our recreation is the result of Expert Prompt Engineering. By controlling every variable—from lighting ratios to textural fidelity—We’ve eliminated the slop and maintained the prestige the brand deserves.

2. High-End Quality on a Lean Budget

The original ad likely cost hundreds of thousands of dollars in studio fees. By using a customized AI production pipeline, we can achieve a similar visual impact at a fraction of the cost.

3. Agility: From Idea to Export in Days

Traditional studios are slow. If you want to pivot a campaign for a seasonal event—like a sudden summer heatwave or a holiday weekend—the studio model can’t keep up. Our workflow allows us to collaborate on your creative ideas and move from concept to high-fidelity export in a matter of days.

Your Ideas. Our Execution.

The most common mistake marketing departments make is thinking they need to buy the tools. But tools are only as good as the person wielding them.

The Arrrival Fallacy In Careers

By Kiran Kulkarni on March 21, 2026

Arrval Fallacy is the misconception that reaching a specific goal or destination will make us happy. In reality, while we do hit a dopamine spike when we hit a milestone, it is always temporary. The “dream car” feels great for a week and then it becomes a means of transport.

As a career coach, almost all of my conversations with my coachees is around promotion. People often describe it as a matter of prestige. They believe the title will validate their worth amongst their peers. Of course, once they get the promotion, their peer group changes and soon enough they need the next title to validate their worth among this new peer group.

We confuse achievement with well-being.

A good question to ask is: How would your daily life change a week after you got your promotion?

Another layer of Abstraction

By Kiran Kulkarni on March 18, 2026

The one usecase that LLMs seem to have absolutely cracked is coding.

They are the final layer of abstraction between programmers and the computer. We have built successive layers of abstraction since the time of the transistors.

We abstracted away transistor code (switch on, off) with the chips, the assembly language, C/C++, and finally to higher level languages with huge libraries. Coding models are the latest layer that abstract away the code beneath and lets us create prototypes by writing code in English.

Naval Ravikant calls vibe coding the new product management.

AI is making us work more

By Kiran Kulkarni on March 17, 2026

I dont think anybody would be surprised with the title of a recently published WSJ article titled “AI Isn’t Lightening Workloads. It’s Making Them More Intense“

The article was based on new research found that we spent MORE time on Emails, Chat, Softwares like HR and Accounting, and LESS time on focused concentrated work. i.e. participants worked faster and harder, but on shallow tasks and constantly switching contexts.

Berkeley Professor Aruna R. is quoted in the article saying “AI makes additional tasks feel easy and accessible, creating a sense of momentum”

It looks as though the productivity gains of AI are being used to get more tasks done, and less to improve the quality of work – and eventually our lives.

SaaS is evolving, not dying

By Kiran Kulkarni on June 10, 2025

There has been a lot of talk during the last two years about the Agentic era leading to the death of Software as a Service aka SaaS. In June 2024, the Salesforce stock slid more than 20% based on its quarterly earnings. It became especially loud recently when Satya Nadella recently stated his views in this interview

The argument is that all SaaS business at its core is a database and an application wrapped around it. The application contains business logic and dashboarding capabilities. A well trained agent based on an LLM should be able to replace the business logic in the future.

Indeed, pure SaaS platforms, where the applications are the classic CRUD (Create, Read, Update, Delete) actions on a database and some visualization built on top – are becoming commoditized. These platforms can be built in google cloud tools in a matter of days.

In theory, one could build an agent, connect it to an existing (set of) SaaS application(s) and use it, and thereby kill the app interface. There are following challenges

Missing Domain expertise: SaaS applications, possess not just data, but also domain expertise. This expertise is built into the complex workflows and business processes. For e.g. there are workflows that determine how a new lead is qualified in a CRM application. In a real automation, the agent will automate the existing workflows everytime a new lead needs to be qualified.

Relying on user prompts: People imagine that an Agent can be provided connectivity to a tool and thereby kill the SaaS interface. However, the product experience would still rely on the user’s prompting skills. We know very well the challenges of asking imprecise or very complex questions to an over eager intern. They will rely on their general knowledge and desire to please you to provide you with great – but unhelpful responses. In some cases one might still require the user to provide login credentials, and then the prompts.

MCP:

I think the more likely approach will be to have agents as native experience within the SaaS app itself. Using MCP protocols, we will build architectures that call on SaaS Agents to automate tasks within that platform for us.

SaaS is a system of engagement built on a system of records. At least in the near future, agents will not replace the system of engagements. It is likely that SaaS agents will offer an improved engagement and automation.

Aftermath of the tech layoffs: How to land job interviews

By Kiran Kulkarni on February 19, 2023

The recent round of tech layoffs has affected a lot of people. There is the emotional stress of being abruptly laid off, coupled with an unfavorable job market. A lot of them have reached out to me recently for help and advice. While the macro economic head winds are real and painful, I see three common errors that one can replace with actions that are more helpful.

Candidates believe that they are at the far end of a long series of hurdles that they need to overcome before they reach the job interview that they deserve. They commonly employ three strategies.

Keyword stuffing: Candidates try to trick the scanner bots by jamming in as many keywords as possible. The potential employer is left wondering what – and how much – is actually true. Of course, every one of the ~884 candidates used the same trick.
Generic and long resume: This applies more to foreign candidates. They believe a 10 page resume in Ariel font size 10 is the best way for the potential employer to appreciate the value that the candidate brings.
Spraying the resume: The candidates feel they need to apply for as many positions as quickly as possible. They apply to anything and everything remotely matching their skillsets; and why not – applying for a job on linkedin takes approximately 3 clicks.

The combination of these tactics is a generic resume. It has something for everyone and gets sent to everyone. It does not work because we are expecting the potential employer to do the emotional labor of finding your suitability for the job.

Before I suggest the alternatives, I would highly encourage the book “What color is your parachute”. I have gifted this book to several of my mentees and friends. I personally found it very useful back in my 20s as I was looking for a job.

The core idea of this book is that if you want to get hired, think like an employer. I would go a step further. Do the emotional labor on behalf of your employer. Here are three alternatives to the above mentioned behaviors.

Network: I do not mean it as a social activity. Sending linkedin invites does not take any effort. Networking means developing peer relations. It takes time. Share something of value with your network. A network of trusted fellow colleagues will cut the SEO process far better than keyword stuffing.
Demonstrate: The resume is an A4-sized visiting card. Adding a cover letter does the emotional labor of matching your resume to the role the candidate is applying for. If one has a tech resume, your github repository is worth more than a thousand words. Link the reddit or stackoverflow contributions that you have made.
Homework: We all work at the intersection of tech and something else. For e.g. I am a cloud architect working with media companies. While the tech sector is laying off, the media sector continues to hire. If you are the candidate, identify the prospective industries, size and roles .

The internet has removed the friction in the application process. However, the employers have raised artificial barriers to cope with the resulting deluge of resumes. The way out is to treat the employers like humans, not an AI model.

Google, ChatGPT and the Innovator’s dilemma

By Kiran Kulkarni on February 5, 2023

In the booming post World War US economy of the 1950s, the American companies like RCA and Zenith made a ton of money selling music consoles – elegant cabinets with integrated radios. It was part of the middle class living rooms. High quality wood and high quality radio.

Sony, a small company back then, licensed the patented transistor technology from AT&T and made small, portable radios. It didn’t compete on quality. It made transistor radios accessible to teenagers for the first time. The product provided Sony with a foothold in the American market from where it could gain market share.

The book Innovator’s dilemma describes this as follows.

The incumbent is focused on existing customers and providing high value products. New companies serve low-value customers with poorly developed technology. There is a lot of trial and error before the technology matures. The innovator’s dilemma is to identify which innovation is going to take off and compete with them.

What does this have to do with ChatGPT?

Google search is the incumbent player. People all over the world rely on it to provide relevant, accurate answers to their questions. Accuracy and relevance are the premium features here, much like the sound quality and wood quality were in the music consoles.

Chatbots based on generative AI are the next wave of innovation. The technology is still in early stages and we have already seen two very famous and public failures

Microsoft Tay, launched in 2016, started spitting out racist, misogynist slurs within 24 hours of its launch
Facebook Galactica, launched in 2022, was shut down within 3 days after embarrassingly inaccurate results

It is dangerous for a dominant player like Google to risk putting something out there without sufficient safeguards around accuracy and relevance. ChatGPT seems to have avoided that embarrassment so far and Google’s dilemma is if the technology is sufficiently advanced. By most accounts it will likely launch its own response soon.

The book Crossing the Chasm is the more relevant one for me here. There is a chasm, a gap between the early adopters and the rest. The early adopters (the ChatGPT users) want something new but the rest want something that is proven.

A company that successfully crossed the chasm is Apple. Apple was not the first one to launch the watch or the headphones. But when they were finally launched, the technology had matured and they matched our expectations of a premium quality product.

Google is now where Apple was around 15 years ago. I don’t worry that it did not out-innovate a newbie. I hope that it delivers its promise on accuracy and relevance when it does.

Why broadcasters should adopt the cloud

By Kiran Kulkarni on October 4, 2022

Digitalisation in media has led to a gravitational shift from broadcast media to OTT (over-the-top) media, such as Youtube and Netflix. However, in the last few years the broadcasters are catching up. Almost every broadcaster today has its own OTT platform. As they modernise their infrastructure, I see a lot of them asking themselves – what is the value-add of the cloud for a company that already has an established and well functioning broadcast infrastructure. This is especially true for teams working in traditional playout centers – the emphasis on changing everything seems unjustified – even irrational

Let’s start with the fundamental question – Why should a traditional broadcaster want to use the cloud, anyway?

Because that’s where the users are:

The users are consuming content over the internet and traditional broadcast technology is not suited to serve that channel.

Additionally, COVID-19 and the emergence of newer technologies enables traditional broadcasters to leverage the cloud better. From a technology perspective, I see the following reasons why a move to the cloud is required.

The cloud will enable you to stay current.

The cloud gives you access to a continually improving hardware stack, reducing the risk of ending up with ‘legacy’ systems in a world where technology change is accelerating over time.

Remote Work Vs. Location Independent Work

Traditionally, the content and the people were in the same location i.e. the office. It was assumed that they cannot be separated. An exception was remote work – when the employees could still access the files remotely (through a VPN).

During Covid, companies were forced to separate the content from the people. They realized that moving the content to the cloud enables remote workers to access it, edit it, even create new content independent of their location. Security and governance are much more granular in the cloud and independent of the network.

Offline editing in the cloud is now possible with newer technologies

Full video editing needs a lot of processing power because of the large data volumes involved in the editing process. It is an essential part of media workflow and the most important reasons people give to keep it on-premise. Traditionally, it has been done by pushing the content to a high-powered laptop and using editing tools on the laptop.

Now, we keep the content and editing tools in the cloud and use newer technology such as PCoIP (PC over IP) to edit it. We combine it with better codecs to stream the data from the cloud location to your computer with minimal latency over the internet.

The average age of the TV viewer is now in the 50s for most industrialised countries. The broadcasters need to move quickly to get where the younger users are.

Leveraging others’ innovation

By Kiran Kulkarni on August 13, 2022

Anybody who has gone through a slide deck on data has heard of three things.

First, data as the new oil.
Second, the value of the data is in the insights. Insights are generated by combining multiple sources of historical data
And third, the amount of raw data generated in the world doubles every two years.

This exponentially growing raw data needs to be stored for insights (and the accidental crash). Usually the most recent data is stored on disks. The storage of data on disks is expensive, so older data is taken offline and stored on magnetic tapes. Data storage on the tapes is cheap, but then it needs to be copied to disks to access it. i.e. data retrieval takes time.

Data tapes at Google

Every company that I have worked with had a library of tapes that contained backed-up and archival data. In google, in 2019, the tape storage system was around 30 exbibytes, or about 34.5K petabytes. It required 21 million magnetic tape cartridges – roughly 23 times the distance to the moon and back. The process of finding the right tape, transporting it to a datacenter and making it available would take several weeks. The tapes had data from Gmail, Youtube, Photos and many other services that google had launched over the years. Given that this data would exponentially increase, storing it on tapes was not a viable option.

The google technical infrastructure team innovated on an entirely new hard drive-based storage system. It was affordable, secure and immutable. Data from any Google service (for example, Gmail) can be restored within seconds. The customers of google cloud can also restore archival data within seconds and not days or weeks.

Avoiding reinventing the wheel

The journey of creating a massive tape storage system and innovating out of it is a journey that every company will need to take on its way to digitalisation. Fortunately, one does not need to reinvent the wheel every time.

The cloud lets you piggyback on others’ innovation.