ai-automation on S Anand

Tacit is just un-instrumented

Mon, 13 Jul 2026 20:43:33 +0800

At The Curzon Hotel, my key card didn’t work. But every time I went to the reception, they’d send a bellboy who would use the same key card, jiggle it a bit, pull it in and out a few times, and the door would open.

Every night. For five nights. I just couldn’t get the knack of it.

I’ve been at the other end of this. People often reach out to me saying, “Anand, this software isn’t working.” Then I go do the same thing they did, and it works. (Sometimes, I just need to watch them do it and it works.)

It’s an intangible skill, I guess.

That gave me some food for thought. This is exactly the kind of skill an AI cannot pick up, right? I mean, jiggling keys, physical world, tacit knowledge, precisely the kind of things that would be AI proof.

So I asked Claude Fable for its opinion. “Can AI pick up the key knack?”

“It already has.” Claude said. Apparently, opening locks is one of the most studied problems in robotics.

The only reason opening the key feels hard to learn is because we didn’t / couldn’t put it in words. But a sensor on his hand would. Tacit is just un-instrumented. Once we measure it, it becomes training data.

As long as something is cheap to try and fast + clear to verify, it doesn’t matter how “physical” it is - we can build a model around it.

	Cheap to try	Expensive to try
Fast + clear to verify	Pottery	Surgery
Slow + vague to verify	Friendships	Mergers

This applies to organizations firms in different ways.

Organization	Automate (cheap to try, clear to verify)	Cut the cost of trying (expensive to try, clear to verify)	Cut the cost of verifying (cheap to try, vague to verify)	Keep human (expensive to try, vague to verify)
Insurance	Photo-based simple claims	Fraud investigations (proven or not) → AI triage of which cases to open	Underwriting rule tweaks (losses mature in years) → early-warning loss indicators	Risk appetite; reinsurance structure
Asset mgmt	Rebalancing, index tracking (tracking error verifies daily)	Large trade execution (implementation shortfall is measured) → execution simulators	Stock picks (skill or luck? takes years) → forecast scoring, attribution	Private-market deals; manager selection
Waste mgmt	Route optimization; robotic sorting	Fleet electrification pilots (cost per route is clear) → route and energy simulation	Recycling awareness campaigns → bin-level contamination sensors	Landfill siting; 30-year municipal contracts
Logistics	Routing, load planning, ETAs	Network redesign, e.g. a new hub (cost-to-serve verifies in months) → digital twin of the network	Driver incentive tweaks (retention causality is murky) → cohort telemetry	Building capacity ahead of demand
Healthcare equipment	Visual defect detection on the line	Clinical trials (clear endpoints, millions per try) → in-silico trials, device digital twins	Hospital sales messaging (committee sales, vague attribution) → pipeline instrumentation	Ten-year platform bets (R&D + regulation + adoption)
Card processor	Transaction fraud scoring (millions of labeled tries a day)	Core platform migration (latency and uptime verify instantly) → shadow and parallel runs	Fee and pricing tweaks (merchant churn is slow, confounded) → churn cohorts	Betting on new payment rails
Scientific publisher	Integrity checks, formatting, metadata	Replicating a paper’s results (re-run the code and data; verdict is clear) → automated re-execution	Desk rejections (did we reject a breakthrough) → track the fate of rejects	Open-access business model transition
Virtual school	Auto-grading; tutoring on known-answer problems	Full course production (completion and scores verify fast at scale) → AI-drafted courses	Engagement nudges (engagement isn’t learning) → better assessment	Accreditation; university partnerships
Physical school	Timetabling, worksheets, admin	Campus expansion (enrolment verifies) → demand modeling	Classroom pedagogy tweaks (education’s replication crisis) → proper assessment	School culture and head succession

Or at a role level.

Role	Automate (cheap to try, clear to verify)	Cut the cost of trying (expensive to try, clear to verify)	Cut the cost of verifying (cheap to try, vague to verify)	Keep human (expensive to try, vague to verify)
CMO	Ad copy variants (CTR verifies in hours)	National campaign launches → test with synthetic consumers, test markets	Brand and content posts (“half my advertising is wasted”) → brand-lift measurement	Repositioning the company
CFO	Reconciliations, close, variance commentary	Refinancing and hedging moves (P&L verifies) → backtests, scenario sims	Forecasts (cheap to issue, never scored) → track accuracy, Brier-style	M&A; capital allocation
CHRO	Policy Q&A, payroll queries	Comp restructuring (offer acceptance, attrition verify in months) → model before rollout	Training programs (nobody knows if they worked) → real skill assessments	Succession; senior hires; culture
CIO	Code with test suites	System migrations and cutovers → staging, canary, parallel runs	Developer productivity tooling (adopted cheaply, impact unclear) → DORA-style metrics	Build-vs-buy platform bets
CRO (Sales)	Lead scoring; outreach drafts (reply rates verify fast)	Enterprise pursuits (win/loss is clear, each pursuit costs months) → rehearse against simulated buyers	Relationship nurturing (coffee now, payoff unclear when) → pipeline telemetry per touch	Key-account and channel strategy

The common pattern here is:

	Cheap to try	Expensive to try
Fast + clear to verify	Automate high-volume	Build simulators
Slow + vague to verify	Capture data	Spend on leadership development

Keep in mind that this is at a task-level, not role-level. A single role may span the entire spectrum of tasks.

RIP, Data Engineers

Wed, 04 Feb 2026 19:22:13 +0800

As AI marches along, another role at risk is the data engineer / database administrator.

(Data scientists are already feeling the heat.)

A common task for data engineers is to analyze SQL queries - to optimize and standardize.

Pavan used Antigravity to analyze 1,500 SQL queries and found:

30% of queries are purely headcount / volume related. Much more than revenue (25%) or engagement (15%). That’s sign of a tactical culture.
70% of the queries are about What happened yesterday? rather than What will happen tomorrow? - again, tactical culture.

Here’s the analysis.

As a next step, he built a “Middle Layer” - intermediate tables that standardize and optimize queries. Instead of 50 fragile tables, the user can query just 3 robust tables that cover 98% of the SQL queries.

For example:

A net_revenue field that standardizes net revenue after adjustments, i.e. SUM(face_value - discount), which is used in 58% of queries. That ensures that Finance (which used to see the GAAP Revenue) and Sales (which used to see the Booked Revenue) are now aligned.
A tickets_sold field that standardizes distinct count of tickets sold, used in 85% of queries, and is a slow computation.

NOTE: Season ticket buyers often bought merchandise as guests (for convenience). Marketing saw these as new customers and spammed them - annoying VIP customers. This standardization created an identity graph - so they can offer discounts instead.

The process, which Antigravity figured out mostly by itself, was to parse the SQL into an abstract syntax tree (AST), extract a set of features, map them into clusters (archetypes), and analyze them to create the middle layer tables.

SQL queries can reveal organizational culture and misalignment - which is cool! But also:

This took a few hours.
Pavan has no data engineering experience.

RIP, Data Engineers.

Fri, 12 Sep 2025 00:00:00 +0000

Slides for my DataHack Summit talk (controversially) titled RIP Data Scientists are at https://sanand0.github.io/talks/2025-08-21-rip-data-scientists/

Summary: as data scientists we explore, clean, model, explain, deploy, and anonymize datasets. I live-vibe-coded each step with DGCA data in 35 minutes using ChatGPT.

Of course, it’s the tasks that are dying, not the role. Data scientists will leverage AI, differentiate on other skills, and move on.

But the highlight was an audience comment: “I’m no data scientist. I’m a domain person. I’ll tell you all this: If you don’t follow these practices, you won’t have a job with me!”

They’re catching on! 🙂

Things you can use e-mail for

Thu, 10 May 2001 12:00:00 +0000

There’s a lot you can do using e-mail, including upload FTP files, create home pages, send snail-mail, translate languages, play games (blank e-mail), send a fax (no subject, “help” in body), track webpage changes, etc. The full details are available at the Accessing the Internet by Email FAQ