A short note on how AI can help

What’s already changed

AI has already changed how you work, and nobody decided it should.

Michael BordeSreejith DasSeptember 2026
Scroll

The habit you dropped

When did you last click through to a website from a search result?

Now most people only read the AI summary at the top and stop.

Nobody was persuaded, nobody ran a pilot. It was simply easier, so it happened.

Pew Research Center, July 2025: users clicked a search result on 8% of searches that showed an AI summary, against 15% without one.

what is generative AI?
AI overview

Generative AI is a type of artificial intelligence that creates new content, such as text, images or code, by learning patterns from large amounts of existing data.

The answer arrives. The links go unread.

Real at the desk, invisible in the accounts

People who use generative AI say it saves them about two hours a week. Ask their boards and nine in ten firms report no measurable impact on productivity.

Both are true. The time comes back to the person, and they use it for something else.

0.0 hours
a week saved by people who use generative AI
Bick, Blandin and Deming, St. Louis Fed, 2025
0 in 10
firms report no measurable impact on productivity
Yotzov and co-authors, NBER, 2026: nearly 6,000 executives

The loop you’re already in

Think about your last two hours at a screen. You took information out of somewhere, did something to it, and moved it somewhere else.

Deciding what to do about it is most of what you’re paid for. The rest is moving information between systems.

LedgerFinance Scheduling systemOperations People systemHR CRMSales Take it outexport, copy, read Do somethingto itand decide the only part that needs you Move it onre-key, paste, sendback into the same systems
LedgerFinance Scheduling systemOperations People systemHR CRMSales Take it outexport, copy, read Do somethingto itand decide the only partthat needs you Move it onre-key, paste, sendback into the same systems
Different systems in every department. The same loop in all of them.

Worked example: the board pack

The fourth working day of the month. The board pack needs a paragraph explaining why marketing is eleven per cent over budget. The actuals are in the ledger, the commitments in the purchasing system, the phasing in a budget spreadsheet, and none of them use the same cost centre names.

The explanation is three sentences. Getting to them takes most of a day.

ONE WORKING DAY exporting fromthree systems matching cost centrenames by hand chasing the one figureonly the ledger has reformatting intothe board pack layout the analysis: ten minutes
ONE WORKING DAY exporting fromthree systems matching cost centrenames by hand chasing the one figureonly the ledger has reformatting intothe board pack layout the analysis: ten minutes
Ten minutes of judgement; the rest of the day moving numbers between systems.

The same task with help

Give an AI tool the three exports and the layout, and let it learn once how the cost centre names map onto each other. It comes back with the sources reconciled, the odd figure flagged, and a draft for you to disagree with.

And next month, when the same question arrives about a different line, the setup is already done.

Ledgerthe actuals Purchasing systemopen commitments Budget spreadsheetthe January phasing AI reconcilesnames mapped once, odd figure flagged You decidequirk, or a problem?
Ledgerthe actuals Purchasing systemopen commitments Budget spreadsheetthe January phasing AI reconcilesnames mapped once, odd figure flagged You decidequirk, or a problem?
Three exports in, one reconciled view out. The judgement stays yours, and moves to the front of the day.

Why nobody fixed it before

Connecting two systems has always been a project: specified, scoped, costed, prioritised, delivered eighteen months later. Anything smaller, a person does by hand indefinitely.

AI has moved that threshold. The person with the problem can now describe it to an AI tool and assemble the fix themselves, often in an afternoon, without waiting for a project.

bigger problems replace the finance systema new CRM reconcile three cost centre schemes reformat the board pack re-key figures that already exist export because systems won’t talk copy it into a third place big enough to justify a project what it takes now
bigger problems replace thefinance systema new CRM reconcile threecost centre schemes reformat theboard pack re-key figuresthat already exist export becausesystems won’t talk copy it intoa third place big enough to justify a project what it takes now
The work didn’t get smaller. The line moved.

Case study: a two-day report in an hour

A one-off client report needed more detail than the reporting system could give. An engineer rebuilt the data by hand and an analyst checked it: two people, two days, every time. Rebuilding the system properly was estimated at three to four months.

An analyst asked an AI model to solve it. It read the public documentation for the data sources and built a connection any AI tool could use; an engineer helped put it live. The report now takes one person an hour, and the same connection answers everyone who asks.

2 days
two people
every request, before
1 hour
one person
every request, now

Let AI do the boring part.You do the interesting part.

Getting started

Pick one repetitive task and try using AI to help. Use the tool your organisation already provides.

Maybe it saves you ten minutes, maybe it doesn’t work the way you hoped, but you’ll adjust and try again.

Before you start

Your organisation has rules about which data can go into which tools, and they exist for good reasons. Find out what yours are: the task you pick should be one you are allowed to do.

You know best

When MIT studied enterprise AI, the pilots that paid off were driven by the people who had the problem, not a central AI function.

MIT, “The GenAI Divide: State of AI in Business”, 2025.

You already know what you want to fix. Pick something small and see how far you get.

If a conversation would help, we’d be glad to have one.

Who we are

Michael Borde

Michael is a software engineer specialising in artificial intelligence and secure, production-scale systems. He spent more than seven years at Amazon Web Services as a full-stack engineer on an AI service team, working across distributed systems, security, computer vision and large language models. His work there included supporting the NHS in scaling its data systems during the COVID-19 pandemic, and applying computer vision to NFL gameplay video.

Since leaving AWS he has focused on independent AI research: how large language models can reason through and carry out complex engineering tasks. He brings an engineering-led approach to AI adoption: finding the valuable applications, building secure systems around them, and fitting them to existing teams and workflows.

BSc Computational and Applied Mathematics; BSc Computer Science, University of Chicago.

Sreejith Das

Sreejith is a technologist and company founder with decades of experience across banking, finance and technology. His projects include business process re-engineering for the London branch of an international bank and straight-through processing for a multi-billion-dollar finance company.

Most recently he was CEO and co-founder of Attestant, a blockchain technology company entrusted with staking billions of dollars of client assets; its software is used by some of the largest firms in the industry to help run the Ethereum network.

BSc Physics with Electronics and Computing, UCL; PhD Applied Mathematics, Birkbeck; postgraduate AI research in reinforcement learning and neural networks, Imperial College.

What’s already changed · September 2026 Built with the help of AI
Turn your phone upright to read
Scroll to continue