Remember the exuberance around Zoom / Teams during Covid? When the consensus was that this is the "only futuristic way" of meeting people. World over, companies touted how they are saving millions of dollars (+ environment) by not travelling for in person meetings. Some companies have gone a step ahead and concluded that employees working virtually is the only way forward. Nearly 3.5 years after Covid ended, Zoom's stock price is down ~85% from the Covid peak (narrative). I am not sure how many people even remember Webex now! Over the same time frame, Mariott hotels (a business presumed to be dead!) is up ~190%! The same analogue can be extended to pairs like Amazon - Walmart, Snowflake - Oracle, Cathie Wood - Warren Buffet and many others, but I am sure you get the drift!!
Today AI might be at the same level of the hype cycle where Zoom was during Covid!! The current consensus is that AI is the "only futuristic" way of working! Like how they concluded that the future work force will be 75%-100% working from home, many CXOs are now concluding that majority of future work force will just be AI agents (AI first / AI native strategy). Remember the Klarna fiasco? For tech companies which mindlessly extrapolated demand, over hired or did sub-optimal M&A during Covid, AI is now a nice excuse to cut the flab! Some companies are going a step ahead and suffixing their names with .ai - sounding cool and getting better valuations!
Bankers / analysts are publishing hundreds of pages of thematic notes on how AI is the next steam engine, electricity, internet etc.! Why not? A capital guzzling sector which will have to raise hundreds of billions (USD) in equity! Momentum chasing funds are madly driving up valuations of "perceived" AI plays. Journalists are amplifying the fear factor of tech led armageddon - the same way they did on five different occasions over last two decades. Coaching institutes are selling AI FOMO to students even in Tier 2-3 towns in India. Although none of these beat an ad someone posted in my society whatsapp group that read - "Gen AI / Python coaching classes for 7 year olds".
Just to be clear, I am a Computer Science Engineer, a programming enthusiast, a tech analyst and more importantly - a calibrated optimist! I have been a big believer and practitioner of AI. Currently, I am closely working with our CTO in driving AI adoption across our company - which is miles ahead of competition on technology adoption. Evaluated multiple horizontal and domain AI solutions in the market, did due diligence with several vendors promising productivity boost in area of investment management.
I conclude that when used with "adequate guardrails", AI is extremely useful in knowledge work in areas like search, summarizing, documentation, media creation, presentations & coding. However, it nowhere comes even remotely close to replacing people (Measuring the applicability of Gen AI). On the contrary, firms will now have to hire more people in areas like data / cloud / setting up guard rails to be able to help with AI adoption. And unlike consumer adoption of AI, enterprise adoption is going to be slow and challenging!
There is something fundamental to AI which leads to these conclusions of mine. AI is just as good as data (quality, quantity & feedback). LLMs & LRMs are pre trained on billions of parameters and peta bytes of data available on internet (redditt groups, free literature, stack overflow, social media++). In simple terms, LLMs / LRMs offer you recycled knowledge or code. Which if applied incorrectly can have adverse amplified business outcomes, beyond just killing creativity / thought of employees. I spoke to 50+ programmers taking help of AI tools like Github Copilot or Cursor for generating or referencing part of their code. They claim when used correctly "in pieces" to suit their design and flow, these tools improved their coding productivity by 25%-30% (v/s using stack over flow or Google to search for the code). However, when they try generating code entirely using these AI tools, debugging takes much longer than usual and the productivity can deteriorate as much as 30%-40%.
Many premier B school students are now uploading case studies on Chat GPT and submitting final answers within minutes - defeating the whole purpose of a case discussion. What is supposed to be a though provoking exercise is being degenerated into thought less summarization (with 95%+ productivity gain though!). If academia and industry don't step in to put "adequate guard rails", we will be creating generations of zombie managers and engineers who are deprived of critical thought (AI usage - critical thinking correlation). Remember the feedback loop into LLMs which could take down the collective intelligence / wisdom of the mankind. And then there is the structural problem of hallucination of LLMs making them unreliable (Why language models hallucinate?).
State of the art LRMs (O3 mini, Claude 3.7-Sonnet-Thinking, Deepseek) still fail to develop "generalizable" problem solving capabilities with accuracy collapsing to zero beyond certain complexities (The illusion of thinking). Surprisingly, most of the research papers challenging the assumption of quick enterprise scale up of AI are put out by firms like Microsoft, Open AI and Apple. Maybe it is their way of cautioning investors / analysts without having to say it on earnings calls! And before anyone else calling out a bubble in AI narrative, Sam Altmanand Mark Zuckerberg themselves did the needful! But who cares?! Tech investors and analysts largely work on stories, narratives and extrapolation like how they extrapolated Zoom and remote work during Covid!!
Robotic Process Automation (RPA) was a similar buzz of the day nearly a decade ago - although the hype levels were much lower. But, it could not live up to its promise largely because of the siloed organizations and data. We have come a long way from there - courtesy cloud, 5G adoption - although technology viability issues still remain. For enterprises to be able to adopt AI at scale, they have to fine tune and retrain LLMs & LRMs on their internal data. Or develop customized Small Language Models (SLMs). This requires organizational redesign and large precursory IT spends on adjacent areas like cloud migration, data streamlining, governance etc. - a tough ask when research points out to 95% of Gen AI pilots are actually failing (The Gen AI divide).
I believe enterprise AI adoption is going to be a slow 10-15 year journey (like internet, e-commerce, cloud) - unlike consumer adoption of Gen AI or Covid led demand push which happened at lightening pace. And it would be employee additive at best and neutral at the worst. Think of an American telco spending more than half a thousand dollars on Customer Acquisition Cost (CAC). It would be naive of them to replace a human agent in their call centers with an AI agent - just to save a few cents per minute of voice call - likely impacting customer retention! This is even after assuming the current cost of AI agent (15 cents / min vs offshore human agents at 5-6 cents / min) will fall dramatically as AI scales up!!
Even now Zoom gives a significant productivity gain over in-person meetings and most of us use it on a daily basis. And will continue to do so! But, just as a secondary option to prioritize and augment productivity in some cases. Not entirely or even partially replacing the human element. And the productivity gain arising out of this is often reinvested in the form of more meetings / additional tasks / ideation. Not necessarily to fire employees. So, this entire hype around AI replacing people - to justify lofty TAM expectations and valuations - need a reality check!! May be Zoom offers a case study to think through!!
Comments