Aleksandar Janca (@datachad) header
Aleksandar Janca (@datachad) profile picture

I help companies run on autopilot in 7 days. co-founder @ https://teapawt.org/t.co/SU6tg9PMw6 & https://teapawt.org/t.co/EfGsbIIwTT

Athens
Joined January 2026
Photo by Aleksandar Janca (@datachad)Photo by Aleksandar Janca (@datachad)Photo by Aleksandar Janca (@datachad)Photo by Aleksandar Janca (@datachad)Photo by Aleksandar Janca (@datachad)Photo by Aleksandar Janca (@datachad)Photo by Aleksandar Janca (@datachad)Photo by Aleksandar Janca (@datachad)Photo by Aleksandar Janca (@datachad)Photo by Aleksandar Janca (@datachad)
An agent that clicks through a website works until the owner says no. An API integration takes longer to get. One of these is a product. The other is a demo.
1
1
38
Before blaming the model for a broken automation, check the boring stuff: → did the input format change? → did a site or API change its rules? → did a credential expire? That's where most production breaks come from.
1
2
68
"Where does the data go?" has two answers for an agent: → the documents it reads → the history of everything it did Most data-residency reviews only check the first.
3
119
Anthropic now lets you run Claude Managed Agents entirely on your own infrastructure. Agents, sessions and history stay on your servers. Model calls go to the provider you pick. Runtime and model are now two separate decisions.
11
14
337
Big companies are splitting on AI agents: → Amazon: blocks outside shopping agents → United: bans agents from booking → Walmart: gateway with written rules → Southwest: flight inventory viewable in ChatGPT Build on the ones that open a door.
4
63
United Airlines now bans AI agents from booking, changing or servicing tickets. New rule in its contract of carriage, Nov 11. Covers travel agencies and booking platforms too. If your automation depends on someone else's website, it has an expiry date you don't control.
5
1
10
155
Agent demos show the 30 seconds it worked. Ask to see the log from the day it didn't.
7
1
11
185
The agents that last are boring. One job. One data source. One owner. One log.
1
3
80
Anthropic, OpenAI and Google took 69% of enterprise AI spend this year, up from 50%, per Menlo Ventures. Concentration is a risk. Keep your prompts and tests portable across 2 providers.
3
6
168
Open weights let you own the model. They don't give you the person who keeps it running. Budget for both.
1
6
158
Sunday check: which automation did nobody look at this week? Either it's working perfectly or nobody would notice if it broke. Find out which.
2
5
76
If your automation needs a person to check it every morning, you automated the work but kept the worry.
2
3
74
Cheapest AI control there is: a daily spend cap. Takes 5 minutes. Saves the bad month.
3
6
79
Agents are now most of enterprise AI usage, per Menlo Ventures. Most small businesses still have zero in production. That gap is the opportunity this winter.
2
2
65
Weekend task: list every place an AI tool can send something on your behalf. Email, Slack, invoices, posts. Then check which ones need a click.
2
1
100
Last week in AI for small teams, 5 lines: → CrowdStrike: a kill switch for agents → Perplexity: agents that message your coworkers → Snowflake: approved definitions for agents → Kimi K3.1 Pro: open weights, read the license → Menlo: AI spend 2.5x over budget
1
2
78
Friday rule: nothing new goes live after 3pm unless it can be switched off from a phone.
3
4
77
Collective raised $100M to run taxes, payroll and bookkeeping for one-person businesses with AI. 30M solo businesses in the US. The back office is where AI pays first. Not the marketing.
4
4
105
Walmart's rules for AI agents: no access to card details, can't change payment methods, blocked if they break the terms. Good template for any agent that touches money.
1
4
98
Walmart lets AI agents shop with spending limits and purchase approvals. Amazon blocks outside agents. Two answers to the same question. Most small sites haven't picked one yet.
4
6
151