Mikhail Baklanov

[Mikhail Baklanov @MikhailBaklanov]

LLMs are a gigantic sledgehammer.

Most enterprise failures are not "the model is dumb."
They are someone using a sledgehammer to crack a nut — no sequencing, no tie-break, no thin deterministic glue around the blow.

The model is not the product.
The orchestration that keeps the hammer from destroying the table is.

[Mikhail Baklanov @MikhailBaklanov]

The naive assumption is that staying in control means reading every line of AI-generated code.

That approach only turns human attention into the bottleneck.

AI changes the unit of control: from the code itself to its observable behavior, constraints, tests, interfaces, and quality gates.

[Uncle Bob Martin @unclebobmartin]

I’m significantly older than you. I started coding in the late 60s. My current strategy is to not read any of the code written by my agents. That’s the only way I can take advantage of their productivity. What I do instead is to surround the agents with extreme constraints. Unit tests, gherkin tests, QA procedures, quality metrics, mutation testing, test coverage, and a plethora of others. In the end, I have very high confidence in the code they produce because they’ve had to run the gauntlet of all of my constraints and tests.

[Mikhail Baklanov @MikhailBaklanov]

For classical software, the deal was simple: what you can describe, you can automate.

For agents, the deal is uglier: what you can validate, you can automate.

A clever generator without a ruthless critic just produces confident work at higher speed.

The interesting roadmap item is rarely the next tool. It is the judge.

[Mikhail Baklanov @MikhailBaklanov]

Most office workers do not ask IT to build them a spreadsheet.

They make one.

AI agents will become ordinary when creating one feels closer to opening Excel than requesting a software project.

[Mikhail Baklanov @MikhailBaklanov]

Evaluation starts with real inputs

We keep trying to evaluate AI systems before we know what real users will ask them.

A synthetic dataset can tell you whether the pipeline works.

It cannot tell you whether the test resembles the job.

[Mikhail Baklanov @MikhailBaklanov]

TL;DR from 150+ off-stage calls: • job = defend truth under board narrative • ‘show’ beats ‘ship’ in budget theater • headcount is a workshop, not a factory

If you live in feed two, you’re not behind. You’re early to the honest problem.

Which one is loudest where you sit?

[Mikhail Baklanov @MikhailBaklanov]

After 150+ conversations with Heads of AI, VPs of Eng, and internal champions, I stopped trusting the public AI feed.

Off-stage, the job is almost never ‘ship agents.’

It’s surviving the gap between board story and operational honesty.

5 notes they almost never say on stage:

[Mikhail Baklanov @MikhailBaklanov]

3/ The people doing the work are not an AI factory.

They’re small squads switching between security questionnaires, data tickets, and Friday’s live demo.

If your strategy assumes a mature platform organization you don’t have, you’re writing fiction.

[Mikhail Baklanov @MikhailBaklanov]

0.8 × 0.8 × 0.8 × 0.8 × 0.8 = 0.33 - average AI agents accuracy

Your AI agent is 80% accurate at each step.

Put 5 steps into one workflow: 0.8⁵ = 0.33.

Corporate users experience end-to-end reliability—not the best number from your pilot dashboard.

[Mikhail Baklanov @MikhailBaklanov]

@politrash попробуй any.do

[Mikhail Baklanov @MikhailBaklanov]

@unavailable_now @anissa_gr блин, не могу найти такие твиты о чем это вообще было?

[Mikhail Baklanov @MikhailBaklanov]

Пеганов: жи есть короче расходимся http://t.co/kig9mpQXuT

[Mikhail Baklanov @MikhailBaklanov]

Аттракцион невиданной щедрости @ Солянка http://t.co/yYyLDo70OT

[Mikhail Baklanov @MikhailBaklanov]

Бесплатный WiFi теперь и в мвршрутках! (@ маршрутка 4) http://t.co/aRcexhxZwe

[Mikhail Baklanov @MikhailBaklanov]

Опубликовано фото @ Le Carrefour http://t.co/Yo5UA7yY3U

[Mikhail Baklanov @MikhailBaklanov]

I'm at Café Kousmichoff (Paris) http://t.co/mCkbS7m7jV

[Mikhail Baklanov @MikhailBaklanov]

I'm at Hard Rock Cafe Paris w/ @sysoeva_dd http://t.co/3fJuND5dri

[Mikhail Baklanov @MikhailBaklanov]

I'm at Hotel France Louvre (Paris, Ile-de-France) http://t.co/5Agm12r12h