Stephan Blasilli

First principles. Systems, scale & operational leverage. Exploring energy, future of work & tech. Tennis strategy.

[Stephan Blasilli @stephanblasilli]

I created a simple 4-step transformation method. Map the business outcome, understand how the work really gets done, pick tech that fits the process, bring the front line along with MVPs. I write about AI in real operations, power markets, and lean.

[Stephan Blasilli @stephanblasilli]

One number from this week: hiring a supertanker from the Persian Gulf to the Far East now costs $1.3M per day. Before the war it was under $50K. When moving the product costs 25x more, logistics is the strategy.

[Stephan Blasilli @stephanblasilli]

I created a simple 4-step transformation method. Map the business outcome, understand how the work really gets done, pick tech that fits the process, bring the front line along with MVPs. I write about AI in real operations, power markets, and lean.

[Stephan Blasilli @stephanblasilli]

@JavierBlas Freight is the same per tanker. The cargo isn't. At $1.3M a day, the grade decides whose margins survive the voyage.

[Stephan Blasilli @stephanblasilli]

@jayvanbavel The tool is neutral. The habit decides. Shortcut use skips the thinking, tutor use builds it. Roll it out without training the habit and you get the shortcut.

[Stephan Blasilli @stephanblasilli]

AI capex needs AI services at 9% of GDP to pencil out. As big as food. Twice energy. Capital gets paid for revenue, not for capacity.

[Stephan Blasilli @stephanblasilli]

@alexandr_wang The word doing the work is 'collaborated'. The machine writes proof sketches, the mathematician checks them. Same split as everywhere: AI drafts, the expert decides.

[Stephan Blasilli @stephanblasilli]

@greg_ip Capex is a bet. Revenue is the proof. 9% of GDP on AI services means AI would be as big as food and twice energy. Capital gets paid for revenue, not for capacity.

[Stephan Blasilli @stephanblasilli]

@SteveRattner @nytopinion The split is not white collar versus blue collar. It is defined process versus messy reality. AI does not take jobs. It takes the repeatable part of jobs.

[Stephan Blasilli @stephanblasilli]

@BrainsAndTennis Player recognition by skeleton. Next level: identify Zverev by the 17 pre-serve bounces.

[Stephan Blasilli @stephanblasilli]

@murtuza_merc If ServiceNow trains your agents on your workflows, ServiceNow owns your process knowledge. Own the process, rent the tool.

[Stephan Blasilli @stephanblasilli]

@emollick AI raises the floor of every deck. The ceiling stays with the thinking.

[Stephan Blasilli @stephanblasilli]

@emollick Well-defined is doing the work in that sentence. The model takes the defined tasks, judgment stays human, and most junior roles are a bundle of both.

[Stephan Blasilli @stephanblasilli]

Tokens are an input. Finished work is the output. The price of a fixed capability level falls about 13x per year (Epoch AI) while token counts keep rising (a16z: 14x since February). Optimize for cost per completed task.

[Stephan Blasilli @stephanblasilli]

@rubengarciaes Retries, tool calls, idle work, coordination overhead. The agent economy has the same waste categories as any factory floor. Process thinking applies directly.

[Stephan Blasilli @stephanblasilli]

@StockSavvyShay A 15% hike on reserved pricing means AWS would rather ration by price than build ahead of demand. Compute scarcity is the story now.

[Stephan Blasilli @stephanblasilli]

@a16z Up 14x since February. Token demand is compounding faster than efficiency gains, which is exactly why the per-task cost matters more than the per-token cost.

[Stephan Blasilli @stephanblasilli]

@timo_rf Four prompts and all your tokens Argon. Honest naming, really.

[Stephan Blasilli @stephanblasilli]

@sandislonjsak Tokens are the visible part of the bill. The full comparison is cost per finished task, where the human side includes salary, handoffs, and waiting time.

[Stephan Blasilli @stephanblasilli]

@antirez Two flat subscriptions beat the API bill until your usage passes a certain point. After that, the meter wins again. It helps to know where your own point is.

[Stephan Blasilli @stephanblasilli]

@carlquintanilla @business Jobs are bundles of tasks, and a management role is mostly coordination. If the AI takes the reporting and the meeting notes, the question is what is left. This only works if you unbundle the job first and check which tasks stay.

[Stephan Blasilli @stephanblasilli]

Five sets are the product, the calendar is the cost.