Notes from my archive
What a note I wrote in the year 2000 got right about the cloud
Twenty-six years ago I sketched an internal system for my executive-search firm. Reading it again today, I find I had described cloud computing, zero-trust security and data lineage — only the wires to carry them hadn’t been laid yet.
I keep a habit of going back through my old notes. It is not nostalgia — it is a way of checking whether the ideas I was chasing actually went somewhere. The other day I pulled out one dated 26 March 2000, titled, with my usual flourish, “Kalchakra — Wheel of Time.” It was meant to be the blueprint for the “Digital Nervous System” of my firm, 3P.
I expected to find a quaint, dated document. Instead I found that, underneath the year-2000 vocabulary, I had specified things the rest of the world would only name and sell years later.
I described the cloud before the wires existed
Here is what I wrote about how our regional offices should work:
No local processing at regional offices. All databases & applications to reside ONLY at HQ. Regions to log into HQ Server to query, process, enter data, send emails, raise invoices, submit offers — anything and everything to be done ONLY on the HQ server, with a comprehensive log of who did what, when, why and for whom.
Strip away the phrasing and that is a thin-client, software-as-a-service architecture — written six years before Amazon Web Services launched, and while Salesforce was barely a year old. The instinct to keep nothing on the edge and have every office simply connect in was correct. What defeated it in 2000 was not the idea; it was Indian bandwidth. The VSAT and dial-up links of that era simply could not carry a “do everything only on the central server” model. I had the destination right. The road had not been built.
The security section reads even better with hindsight. I asked for fingerprint recognition on every keyboard, coupled with voice recognition, on top of individual passwords; for a person’s credentials to be auto-deleted the moment they left the firm; for no floppy, CD or copying device anywhere; and for HQ to be alerted automatically the instant someone tried to attach such a device. Today we would call the first part multi-factor and biometric authentication, the second part zero-trust, and the third part Data Loss Prevention — a product category that did not even have a name back then.
The business decision I got right — and the irony in it
The same file contains a long, honest essay on our business model. By 2000 most Indian jobsites were, in effect, selling electronic media space — newspaper logic ported to the web. Anyone could see the content and act on it independently of the site. We did the opposite: we kept the identity of jobseeker and recruiter hidden from each other until the recruiter accepted our terms. The platform could not be bypassed.
That was a genuinely sound strategic insight — it is the same logic that protects any marketplace from leakage. But here is the irony I can now admit: the model I diagnosed as weaker is the one that won the Indian market. “Sell the media space” was capital-efficient and frictionless to scale, and the firms that adopted it had the venture funding I openly noted we lacked. My model was more defensible per transaction and far harder to grow without deep pockets. In one memo I had identified both my edge and the structural reason it would struggle. With hindsight, that self-awareness was worth more than being “right” would have been.
A few smaller things that aged remarkably well
Our data-capture module had 97 fields, and my engineer had promised that a consultant could build a unique Excel sheet by picking any number of fields arranged in any order. That is a no-code, faceted report builder. Around it I had written strict rules — which fields are mandatory, which may be edited, which must never be overwritten, and a demand to capture who made the original entry, when, and from what source. That is data lineage and governance, the precise thing everyone now frets about in the age of AI training data.
And my favourite line, on getting senior executives onto the database: consider letting them phone in their resumes by IVRS — they will not sit and type, but they are used to dictating. That is the voice interface, roughly twenty years early.
| What I wrote in 2000 | What it was later called | When it arrived |
|---|---|---|
| No local processing; all data & apps reside only at HQ; regions log in to work | Cloud computing / SaaS / thin client | AWS 2006; mainstream 2010s |
| Fingerprint + voice + password; credentials auto-deleted on exit | Multi-factor & biometric auth; zero-trust | mid-2010s |
| HQ auto-alerted if anyone attaches a copying device; no removable media | Data Loss Prevention (DLP) / endpoint security | mid-2000s onward |
| “Comprehensive log of who did what, when, why, for whom” | Audit trail / data provenance & lineage | now central to AI governance |
| Simultaneous multi-region video conferencing replacing physical meets | Remote work / video collaboration | 2020 |
| Branch attendance integrated into central payroll | Geo-tagged workforce systems (my own MAD framework) | ongoing |
| 97 fields; build any Excel by selecting any fields in any order | No-code / faceted custom report builder | 2010s onward |
| Mandatory / editable / immutable field rules; capture origin of every record | Data governance & lineage | now |
| “Phone in your resume by IVRS — executives dictate, won’t type” | Voice interfaces / speech-to-text capture | 2020s |
| Non-member database built from annual reports, directories, cardholders | Passive-candidate sourcing (what LinkedIn monetised) | 2003 onward |
Where it did not travel — and what I’ll own
Two honest qualifications. First, my “non-member database” — a passive-candidate pool scraped from annual reports, directories and the like — was the right instinct, but the same memo names its fatal flaw: it decays. I warned my staff that without constant editing the records would go obsolete and embarrass us. The acquisition of data was scalable; its maintenance was not. That tension — easy to gather, brutal to keep true — is unresolved even today.
Second, the internal architecture was, by today’s lens, heavy on monitoring my own staff: biometrics everywhere, no local autonomy, everything logged. In 2000 that read as discipline. In 2026 it sits squarely inside the live debate over employee surveillance. The very design that looks visionary on the cloud and zero-trust axis looks contested on the worker-autonomy axis. I’d be the first to say so.
What strikes me most, reading it all together, is the continuity. From this year-2000 nervous system, to my MAD proposal for a geo-tagged national workforce, to IndiaAGI today — I appear to have been describing one continuous nervous system for a quarter of a century, simply at ever-larger scope. The vocabulary keeps changing. The instinct, it seems, has not.
— from the notebooks of Hemen Parekh, who has been writing tomorrow’s memos a little too early since 1999.
