I Audited 20 Years of My Career Against AI. Here's What I Found. AI's Impact on Jobs in India

Mahendra Rathod
7 hours ago
23 min read

Illustration of a woman studying a career timeline scroll while a blue AI figure points at a glowing laptop; books, plant, mug. — If AI existed when you started your career, would your job still exist today?

AI Impact on Jobs India: A 20-Year Career Audit

The question of AI's impact on jobs in India is not abstract anymore. It is arriving role by role, function by function, in companies exactly like the ones I worked in. This is what it looks like from the inside.

I graduated from IIM Bangalore in 2006. I had a BE in Mechanical Engineering before that. I was reasonably smart, worked hard, and had decent analytical skills.

I spent the next 18 years working across six industries — IT services, retail, BPO, e-commerce, health insurance, and startups. I did everything from preparing RFP documents at a large IT company to running operations for 1,200 people across 20 locations.

A few months ago, I started using AI tools seriously. Claude, ChatGPT, Gemini. The more I used them, the more one question kept coming back to me.

If these tools had existed when I started my career — what would have happened to my job?

Not in a theoretical sense. Role by role. Year by year. Honest answer.

So I did the audit.

What I found was uncomfortable in some places and reassuring in others. But more than anything, it was clarifying. Because the pattern that emerged from 20 years of work tells you something important — not just about my career, but about yours.

If you are a senior professional today — 35 to 50 years old, working in a mid-to-large organisation — at least some of what I did in my career looks like what you do now. That is the point of this exercise.

Let's go role by role.

The Clerk with an MBA

Cognizant Technology Solutions, 2006–2008

My first job after IIMB was at Cognizant, in their Life Sciences consulting group. My designation was Associate Consultant. Sounds impressive. The actual work was not.

I was a Business Analyst. In Cognizant's world, a BA was the person who sat between the client and the tech team. The most common task was responding to RFPs — Requests for Proposals. Large IT deals worth millions of dollars.

Here is what that actually looked like. An RFP would come in. There was a standard format — 75% of the content was already written and sitting in a shared folder somewhere. My job was to coordinate with the solution architects, the finance team, the pricing team, the subject matter experts — collect their inputs, stitch everything together into a Word document, get it approved by the project manager, and send it out.

I was a well-educated copy-paste operator.

The other part of my job was MIS. Monthly reports on team productivity, billable hours, attrition, timesheets. I would pull data from multiple sources, clean it in Excel, analyse it, and put it into a PowerPoint. Then send it to the project leadership or the client.

Tools I used: Excel, PowerPoint, and Google for occasional research. That was it.

Now here is the honest question. Could AI do this today?

Yes. Completely.

One person with Claude can manage five RFPs simultaneously. The boilerplate drafts itself. The inputs from various teams get synthesised in minutes. The final document is better written than what most BA teams produced. The MIS reports? A BI tool connected to the data source generates those automatically, without a person touching them.

The BA role as it existed in Indian IT in 2006 does not exist today. I want to be clear — the work was real and it took real effort with the tools available then. But the tools have changed completely.

There is also a bigger point here. Cognizant was not hiring BAs because the work genuinely needed that many people. Indian IT in 2006 was growing at 30–40% every year. The business model was simple — hire large numbers of credentialed people in India, bill their hours to clients in the US and UK at a margin. The more people you had, the more you could bill. BAs existed because the billing model rewarded having them, not because the work demanded them.

AI does not just eliminate the BA role. It eliminates the business logic that created it.

My boss at Cognizant once got me an L1 visa. He wanted me to go to the US as a Technical Writer. I had no idea what a technical writer did. I told him I had done an MBA and wanted to work on analysis, consulting, strategy. He said this is the opportunity — go now, figure out the rest later.

I said no and resigned.

In my farewell meeting, he told me I was going to "sell biscuits." (I had told him I was joining a footwear retail company. He confused the two. I did not correct him.)

I think about that conversation sometimes. If I had gone to the US as a Technical Writer, I would have spent years doing work that no longer exists. Saying no to that visa was one of the better decisions I made, even though it did not feel like it at the time.

AI Verdict: Role eliminated. Not because the person was not good enough. Because the role itself was always going to be automated — AI just made it happen faster.

The Analyst

Reliance Retail (Footprints), 2008

After Cognizant, I joined Reliance Retail's footwear business — Footprints. Eleven stores, around 100 people, headquarters in Bangalore.

I was the only analyst for the entire business unit. My reports went directly to the CEO and the leadership team.

That was a significant step up from Cognizant. But the nature of the work was similar in one important way — it was still largely data compilation and analysis. I got structured data from their ERP system, ran pivot tables, sliced it different ways, and built standard reports. Weekly dashboards, monthly KPI reports, sales per square foot, inventory ageing, brand performance, size curve analysis.

The CEO once told me: "You are the judge of the business. I am the police."

He meant that my job was to present facts — what the data said, clearly and honestly. His job was to act on it. I liked that framing. It gave me clarity on what I was there to do.

The standard reporting work — I have to be honest about this — is almost entirely automatable today. A modern BI tool connected to the ERP produces live dashboards without a person building them every week. What took me several hours in Excel is now a scheduled refresh that happens automatically.

But there was another layer to the work that is more interesting.

For special projects, I had to go beyond the data. One example: we found that stores in Punjab were selling significantly more in larger shoe sizes. Stores in Ahmedabad consistently responded better to promotional offers than stores in other cities. These were behavioural patterns by region — and acting on them meant stocking differently, pricing differently, running different promotions in different markets.

These insights came from the data. But someone had to think to ask the question first. The data does not tell you to look at size distribution by region. A person notices something, gets curious, and decides to investigate.

Here is the honest complication though. Today's AI does not just answer questions you ask it. It surfaces patterns you did not think to look for. A machine learning model on that same ERP data would likely have found the Punjab size curve before I did. And it would have found ten other patterns I never noticed.

So even the insight layer — the part I thought was the most human — is under pressure.

Nine months in, the 2008 recession hit. Reliance had expanded aggressively and ran out of runway. The business contracted sharply. My role went with it.

AI Verdict: Standard reporting and dashboards — fully automated. Regional behavioural insights — under pressure from pattern recognition tools. What survives is the person who frames the right question. But even that window is narrowing.

The Strategy Person

Aditya Birla Minacs — CEO Office, 2008–2012

Minacs was a $450 million BPO company. 22,000 people across India, the US, Canada, and Manila. 80% of revenue from North America.

I joined as the person who supported the CEO directly. No fancy title for it then — today it would be called Chief of Staff.

The first three years were a mix of two very different types of work.

The first type was familiar. Company-level MIS, board meeting decks, investor presentations, planning and budgeting reports. Same DNA as my previous roles — collect data from across the organisation, compile, analyse, present. The difference was the audience. These reports went to the board, to investors, to the chairman. The margin for error was near zero.

This work is heavily automatable today. AI generates board decks from structured data. Financial summaries, variance analysis, budget forecasts — these are not creative tasks. They follow a logic that machines handle well.

The second type of work was different.

I was involved in M&A. Not as a coordinator — I built the analysis myself. Target identification, synergy assessment, financial projections. No templates. I would read the target company's annual reports, look at their locations, their people, news signals, competitive positioning — and build the financial case from scratch in Excel.

AI would be a serious force multiplier here today. Reading annual reports, synthesising signals across competitors, mapping geographic presence — Claude does all of this faster than any human. Target identification specifically is something AI handles well.

But there was one part of M&A that no model touches. Whether the founder of the acquisition target would stay post-merger. Whether the leadership team would survive integration. Whether the cultural fit was real or just looked good on paper. Those assessments came from conversations and judgment — not spreadsheets.

The third part of the role was program management. Post-merger integrations, organisation-wide KRA setting, attrition reduction initiatives. I was the person from the CEO's office assigned to work with functional leaders and report progress back.

This sounds administrative. It was not. These leaders were senior, experienced, and had their own agendas. Getting them to move required credibility — and that credibility came entirely from my proximity to the CEO. I was not there because I knew more than them. I was there because the CEO trusted me to give him an accurate picture of what was actually happening, without distortion.

That trust layer is not something AI replicates. A CEO cannot send Claude into a room with a resistant functional leader to sense whether the person is genuinely committed or just performing compliance. That requires a human — specifically one whose judgment the CEO believes in.

AI Verdict: MIS, decks, financial modelling — heavily automatable. M&A analysis — AI accelerates it significantly, but the people judgment stays human. The program manager function — depends entirely on trust and proximity. That role only exists if the CEO believes in you. AI cannot build that.

The Internal Consultant

Aditya Birla Minacs — Business Intelligence Unit, 2012–2014

After three years in the CEO's office, I moved to lead a unit we called Business Intelligence. A team of ten analysts. Our job was internal consulting — going into business units and finding ways to improve profitability, reduce costs, maximise revenue.

I want to tell you about one specific project. Because it is the best example I have of where AI ends and the human begins.

Minacs had a large client in India. One of our biggest. I was looking at two reports that existed independently of each other — billable hours and paid hours. Nobody had thought to put them together.

I merged them.

What came out was striking. Only 30% of the hours we were paying for were actually being billed to the client.

That number alone is not enough. A gap like that has many possible explanations. So I went deeper. Employee level data. I started looking at where the non-billable hours were going.

What I found: the business unit had far more trainers than required. Training was being scheduled during billable hours, pulling productive staff off the floor. There were too many non-billable overhead roles — layers of management that the contract did not support. And on top of that, the client was making frequent changes that disrupted workflow and reduced productive time.

I did not stop at the data. I conducted interviews. I spoke to people across the unit to understand the why behind the numbers. The data told me what was happening. The conversations told me how it had been allowed to go on.

Then I built the complete picture — not just an Excel file with a gap highlighted, but a full narrative. What the problem was, why it existed, what it was costing the business, and what needed to change.

I presented it to the leadership team knowing that some senior people would lose their jobs as a result of this finding.

They did.

The business EBITDA moved from deeply negative to significantly positive. I received a bonus and a promotion.

Now — what would AI have done differently here?

The gap itself — merging two datasets and flagging a 70% non-billable rate — that is a straightforward query today. Any analyst with a modern BI tool finds that in minutes, not weeks. AI would have surfaced the number faster than I did.

But the hypothesis that led to the query — the instinct to merge those two specific reports — came from a person who was paying attention and asking uncomfortable questions. That still requires a human.

More importantly: the investigation. The interviews. Reading whether a manager's explanation was genuine or defensive. Understanding the politics of why a problem had been ignored for so long. Building a presentation that told the complete story rather than just exposing numbers. And then walking into a room of senior leaders and presenting findings you know will end careers.

AI does not do any of that.

This is what I would call orchestration. Hypothesis, investigation, interviews, synthesis, presentation. Each step requires judgment. The data is only one input in a much larger process.

I also built something during this period called the "Protect P&L" framework. It was an end-to-end business model for a contact centre — contractual metrics, actuals, gaps, billing rates, staffing ideals versus actuals, key assumptions. If a business unit head entered their data, they got a complete picture of how their operation was performing versus how it should be performing. Not budget versus actual — ideal versus actual.

AI could build a version of that framework faster today. But applying it across different business units — adjusting assumptions when business dynamics change, knowing when a BU head is gaming the inputs — that still requires someone who understands the organisation.

The honest AI assessment for this phase: AI compresses the timeline significantly. What my team of ten did, three people with AI tools could probably do today. But the role does not disappear — it concentrates. Fewer people, higher judgment, higher stakes.

AI Verdict: Data investigation and pattern finding — AI is faster. Framework building — AI accelerates it. But hypothesis creation, the investigation layer, the interviews, and the courage to present what you find — those remain human. This role shrinks. It does not disappear.

The Operator

Dell, 2014 — Sellerworx, 2015–2016

Dell

After Minacs, I joined Dell. The reason was simple. At Minacs, I had spent years identifying problems and recommending solutions. I wanted to actually implement something for once.

It did not work out that way.

My role at Dell was to forecast call volumes — how many customer service calls would come in, by issue type, across different time periods. The method was mechanical. Take past data. Project it forward. No causal logic. No understanding of why call volumes moved the way they did. Just extrapolation.

I had almost no work. And the role had no room to grow into anything more interesting.

I left after six months.

AI verdict on the Dell role: a basic forecasting model does this better than a team of humans. This was never a meaningful job. It was a process that should have been automated years before I arrived.

The more important thing about Dell is what it confirmed. Every time a role tried to reduce me to process execution, I left. Cognizant wanted to make me a technical writer. Dell wanted to contain me in mechanical forecasting. Neither was the right fit. That instinct — to move toward complexity and judgment, away from process — turned out to be the right instinct for an AI world, even though AI was not the reason for it at the time.

Sellerworx

Sellerworx was a startup founded by a friend. The idea was straightforward: help Indian sellers get onto online marketplaces like Amazon, Flipkart, and Myntra. There were around 18 such platforms at the time.

The opportunity looked large. India has crores of MSMEs. Even if 1% of them wanted to sell online, the market was significant. Most small Indian sellers did not speak English, did not understand marketplace terms, did not know how to manage listings or handle returns. We would solve that — through technology, cataloguing, imaging, account management, and consulting.

We started in a 1BHK office near Domlur in Bangalore. Four or five people. We raised $1 million from a venture fund with a promise of another million to follow. We hired, expanded, onboarded sellers, built software.

Within six months I had built a team of 100 people and we were managing significant GMV across 1,000+ seller accounts.

But the business model had a fundamental problem. And I only fully understood it after spending three months on the ground in Surat.

Surat is one of India's largest textile and saree wholesale markets. We wanted to bring these sellers online. I went there to understand what they needed so we could build the right product for them.

What I found was not what I expected.

These sellers were not struggling. Their offline business was working well. And when I sat with them and walked through what selling online would actually mean for their day-to-day operations, it became clear why they were not rushing to make the switch.

Offline: almost no returns. Payments same day or within a few days. Fixed customers who came back every season. No discounts beyond normal seasonal variation. Manageable ups and downs.

Online: no guarantee of the buy button — the large in-house sellers on these platforms always got preference. Catalogue copied by competitors within days of listing. Fake orders. Payment after 15 to 30 days. Limited seller protection. Returns that could eat into margins significantly.

The offline model was not being missed by these sellers. It was being rationally avoided.

No dataset would have told me this. This insight came from being physically present. From sitting across a table from a saree seller in Surat and watching his face when I explained how marketplace returns worked.

We eventually sold Sellerworx to Capillary Technologies. The business model had hit a wall — marketplaces were building their own seller services in-house, our smaller sellers were not getting the buy button, and the second round of funding did not come.

But the Surat lesson stayed with me.

AI can process data about sellers. It can analyse marketplace trends, model unit economics, identify which product categories are growing. All of that, faster than any team.

What AI cannot do is go to Surat. A founder still has to get on a train, sit in a wholesale market for three months, and find out what the data was never going to tell them.

AI Verdict: Operations, cataloguing, account management, seller onboarding — significant automation possible. The 100-person team probably becomes 30–40 with today's tools. But the ground truth discovery that changes your entire business assumption — that still requires a human being physically present.

The Business Leader

Loyalty & Gifting Startup, 2016–2019 — Medi Assist, 2019–2020

This is where the nature of my work changed most fundamentally.

Every role before this had a significant individual contributor component. I was doing the analysis, building the models, preparing the decks. From this point, my primary output was decisions, direction, and judgment — delivered through other people.

Loyalty & Gifting Startup

I joined as VP Strategy and Business Operations at a Bangalore-based rewards and gifting platform. Eventually moved to Chief Customer Officer. Around 70% of the company reported into me across five functions — Vendor Management, Operations, Customer Support, Account Management, and Business Finance.

What the role actually required was problem solving at speed, across five different functions, simultaneously, in a company small enough that you could not delegate the thinking. Every day had a different crisis. A large vendor integration failing. A key account escalating. A pricing decision needed before end of day. A team leader not performing. A legal document needing review.

I was also advising the founders on strategy, sitting in on investor meetings, helping with fundraising, vetting contracts, and driving business finance. In a large company, these would be five separate roles. Here they were all mine.

I did a lot of root cause analysis. When something broke, I went into it — not just assigned it to someone. Identified the problem, provided the solution, and made sure the process changed so the same problem did not come back.

Could AI have helped here? Yes, significantly. Research, analysis, financial modelling, drafting documents, summarising data — all of that would have been faster with today's tools. A two-week investor deck exercise becomes two days. A pricing model that took three days of Excel work gets done in an afternoon.

But the core of the role — the judgment calls, the founder advisory, the client relationships, the reading of situations that had no clean data attached to them — that was not an analytical function. It was a human one.

Medi Assist

Medi Assist is India's largest Third Party Administrator for health insurance. They process around 4 million claims a year covering over 2 crore lives.

I joined as Business Head and later became Head of Operations. I was directly responsible for 1,200 people across 20+ locations.

The domain was new to me. Health insurance claims processing is a 25-year-old structured process — policy issuance, document collection, claim tabulation, medical adjudication, fraud detection. Most of it was manual. Most of the people doing it had been doing it for 15 to 20 years.

The hardest part of this role was not the operations. It was getting experienced people to follow a new direction.

My AVPs — the leaders running each line of business — were mostly doctors or senior insurance professionals. They had deep domain expertise I did not have. They had been in the organisation far longer than me. They had no particular reason to move quickly for someone who had just arrived.

What I did was simple. I got a data analyst and asked him to start publishing key operational numbers by line of business, every hour. TATs, claim volumes, error rates, pending files. Then I started asking questions in our daily reviews. Why is this number not moving? What happened here?

Initially there was resistance. That is normal. People do not like being held accountable for numbers they were not previously measured on.

But I also did something else. I stood behind my team when internal pressures — from client-facing functions that wanted their escalations prioritised — tried to push operations into shortcuts. I gave promotions and salary hikes to people who deserved them. I introduced an incentive mechanism for claim processors — accurately process above a certain number of claims per day and you earn Rs 50 per claim, with a percentage going to your manager.

I went into the problems with them, not above them.

Within a few months the AVPs started supporting the new direction. TATs improved significantly for cashless claims. Error rates came down. We brought substantial automation to the policy enrolment function using RPA — reducing the team size considerably within three months.

Now — what does AI do to this role?

The RPA implementation I drove at Medi Assist is a preview of what is coming across the entire claims processing chain. Policy enrolment was manual — field by field data entry from paper documents and PDFs. RPA automated a large portion of it. The next step, which is already happening in the industry, is AI doing the first-pass adjudication of claims — reading the policy, the discharge summary, the tabulation, and making a recommendation on what is payable. The doctor-adjudicator role does not disappear but it changes. From processing every claim to reviewing edge cases and exceptions.

Fraud detection, error identification, document classification — all of these are moving toward AI fast.

But the operational transformation I described — the hourly dashboards, the accountability culture, the incentive design, standing behind the team against internal pressure — none of that is an AI function.

The AVPs did not follow me because I knew more about health insurance than they did. They followed me because I showed up, went into the problems with them, and made it worth their while to perform. That is not something you can automate.

AI Verdict: Claims processing, enrolment, fraud detection, document management — significant automation underway right now, not in five years. The operational leadership layer — building trust with resistant domain experts, designing incentives, creating accountability — that remains human. The more people-intensive the role, the more AI-resistant it is.

The Founder and Mentor

PICO, Vananam, TrekNomads, 2021–present

After Medi Assist, I took a three month break. I looked at doing a PhD in behavioural economics. Decided against it. The job market was slow. So I started doing what I knew — helping startups with business models, investor decks, and fundraising.

That consulting work eventually became PICO — a pan-IIM consulting organisation helping startups raise funds and build businesses. I also took on a director and minority shareholder role at Vananam, a business conglomerate building reward and gifting solutions for B2B clients.

In this phase, AI has already changed how I work. Meaningfully.

A business model that used to take a week now takes two days. An investor deck that needed three rounds of revision gets done faster and cleaner. Research that required hours of reading is compressed into minutes. I use AI tools every day.

But here is what has not changed.

The founders I work with are not paying for decks. They can get a decent deck made cheaply now. What they need is someone who has seen enough businesses fail and succeed to tell them which assumption in their model is the one that will kill them. Someone who can look at their go-to-market and say — this is the Surat problem, and you will only find out you are wrong after you have spent six months and a lot of money.

That pattern recognition comes from 20 years of being wrong in expensive ways. AI does not have skin in the game. A mentor does.

TrekNomads — the trekking community I co-founded — sits outside this analysis entirely. Nobody is automating the experience of standing at 14,000 feet watching the sun come up over the Himalayas. Some things are just human.

AI Verdict: The execution layer of consulting — research, modelling, deck production — significantly compressed by AI. The judgment layer — knowing which assumption will fail before it does — that compounds with experience. AI makes you faster. It does not make you wiser.

What the Audit Reveals

Twenty years. Eight roles. Here is the honest summary.

A few patterns emerge when I look at this honestly.

The first pattern: the earlier the role, the more completely AI replaces it. The Cognizant BA role and the Dell forecasting role are gone — not partially disrupted, fully eliminated. Not just by AI either. Tools like Tableau and Power BI had already started automating the MIS and dashboard work years before AI arrived. AI just finished the job.

The Reliance Retail analyst role started disappearing the moment companies connected their ERP systems to BI tools. Live dashboards replaced weekly Excel reports. That happened through the 2010s — before ChatGPT existed. AI adds the next layer — pattern recognition that surfaces insights the analyst never thought to look for.

The second pattern: the roles that involved coordinating people and information are partially automated but not eliminated — the Minacs CEO office, the program management work. AI drafts the deck. AI summarises the data. But the person who sits between the CEO and the organisation, translating intent into action and reporting reality without distortion — that role still needs a human. For now.

The third pattern: the roles that required physical presence or people leadership are the most AI-resistant. The Surat discovery could not have come from a dataset. The Medi Assist operational turnaround required someone to walk into a room of experienced doctors and earn their trust over months. These are not analytical problems. No intermediate technology — not Tableau, not RPA, not AI — changes what these roles require.

The fourth pattern: AI compresses teams, it does not always eliminate roles. The Minacs BI unit had ten analysts. With today's tools, that work gets done by three. The Sellerworx operations team of 100 becomes 30–40. The Medi Assist enrolment team reduced considerably with just RPA — before AI entered the picture. AI will compress further. But someone still needs to ask the hypothesis, run the investigation, and present the findings.

The clearest conclusion from this audit: AI replaces the information layer of almost every role. Compiling, reporting, modelling, drafting, analysing — these are now cheap and fast. What remains is the judgment layer. Framing the right question. Investigating what the data cannot tell you. Leading people through change. Making the call when the answer is not in the spreadsheet.

The further your work is from the judgment layer, the more exposed you are.

My personal audit is not an isolated data point. The World Economic Forum's Future of Jobs Report 2025, which I covered in detail earlier on this blog, surveyed over 1,000 employers across 55 economies and arrived at strikingly similar conclusions. The fastest declining roles by 2030 — bank tellers, data entry operators, cashiers, administrative clerks — are precisely the information assembly roles my early career was built around.

The fastest growing skills — analytical thinking, resilience, leadership, the ability to work with AI — are exactly what the later roles in my career demanded. And the report's finding that 39% of workers will need reskilling by 2030 is not a distant warning. For many people reading this, that number applies right now, not in five years. The report gives you the macro picture. My audit gives you a way to apply it to your own career, role by role.

🔗 Read my full breakdown of the WEF Future of Jobs Report 2025 here

So What Do You Do?

I am not going to give you generic advice about "upskilling" or "staying relevant." You have read enough of that already.

This is what I would do if I were looking at my career today through the lens of this audit.

Start using AI tools now — not to protect your job, but to find out what your job actually is.

Take the parts of your work that AI can do and let AI do them. What is left after that — that is your actual value. Most people have not done this exercise. They are still doing the full job manually, which means they are spending significant time on work that is either already automated elsewhere or will be soon.

Use the time you recover to go deeper into the parts AI cannot touch.

Move toward the judgment layer deliberately.

Every role has two layers. The information layer — compiling, reporting, analysing, drafting. And the judgment layer — framing the problem, investigating what the data misses, making the call, bringing people with you.

AI is eating the information layer fast. The judgment layer is where the value is concentrating.

If your current role is mostly information layer work — and you will know honestly if it is — start moving. Get closer to the decision. Get closer to the client. Get closer to the complexity. Volunteer for the project where the answer is not obvious. The further you are from the decision, the more exposed you are.

Invest in interpersonal capital. It is the only thing that compounds in a way AI cannot replicate.

The Medi Assist AVPs did not follow me because I knew more about health insurance than them. They followed me because I was present, fair, and genuine. I went into the problems with them. I stood behind them when it was easier not to.

That kind of trust is built over time, in person, through consistent behaviour. It does not have a shortcut. And it does not have an AI equivalent.

Your network, your reputation, your relationships with people who have seen you perform under pressure — these matter more now than they did five years ago. Not less.

And if you are currently in a role that looks like my Cognizant or Dell role — be honest with yourself about it.

The role may not exist in its current form in three years. That is not a reflection of your capability. It is a reflection of what the role is. The earlier you see it clearly, the more time you have to move.

Closing

I started this audit expecting it to be uncomfortable. It was, in places.

The first few roles were hard to look at honestly. I was doing work that tools can now do better. That is just true.

But the audit also showed me something I did not expect. The work I am most proud of — the billable hours finding at Minacs, the three months in Surat, the Medi Assist turnaround — none of it was work AI would have done well. All of it required me to show up as a person, not a processor.

That is not a coincidence. It is the pattern.

If you are reading this and wondering whether your job is safe — I understand the anxiety. But I think you are asking the wrong question.

The better question is: in your current role, what would be left if AI did everything it could do today? If the answer is enough to build on — build on it, and build fast. If it is not — start moving now. Not next year. Now.

The audit does not lie.

Happy Reading!