We used to count the clicks it takes to get a job done on websites and apps. The more clicks, the worse the user experience. The same applies to agentic experience: the more round trips an agent needs, the worse the experience for agent and user.
I reviewed things we added in @apify in recent months to help users build Actors, and decided to share them also here as a quick recap.
Please, anyone building on Apify, repost it, and feel free to provide us with feedback, or things you would like. You can have influence on these things!
Apify CLI
- apify api: call the Apify API straight from the terminal
- apify actors search: find Actors in Apify Store without leaving the terminal
- Agents can run the CLI unattended: --yes on every prompt, JSON output, wait commands for builds and runs, and the Actor's exit code from apify run
- apify create --source: create an Actor from a GitHub, GitLab or Bitbucket repo, with automatic builds
- apify push works in monorepos, so one repo can hold many Actors
- Multi-account login: switch accounts with one command, with credentials stored in the OS keyring
Apify MCP server
- Your agent can create, build and deploy an Actor through MCP, then read the build logs
- It can also search, read and write through the Apify API, with guards that refuse unsafe requests
Agent Skills
- Actor development skill: guides the agent through building an Actor, including its schemas and README
Actor templates
- AGENTS.md in every template, so a coding agent knows how to work on the Actor
Apify Console
- Create an Actor from a GitHub, GitLab or Bitbucket repo in a wizard, with builds on push set up for you
Agent evals night was a banger 🔥
@benswerd driving Claude Code crazy by making it play cookie clicker was my fav 🍪
Main takeaway: benchmarks SUCK and get gamed. The best agent evals are ambiguous tasks, like sending them to play StarCraft against each other
We started shipping the Nvidia H100 in 2022.
A year ago it rented for $1.96/hour.
Today: $3.10.
+58%, on a chip that was supposed to be obsolete by now.
So much for GPU depreciation.
One of the finest and funniest event I’ve been to in SF recently.
The highlight of the eve: @benswerd from @freestyle_dev revealed that you actually can drive claude code crazy if you send it to play “click the cookie”
In 2026, 10 min talks is the max length human with a phone in his hand can focus on. So shoutout to @apify for setting up this event that is almost an irl tech version of tiktok dedicated to evals lol
Reality of our days:
1. I am sitting on a set of vulns (4+) that allows me to remotely take over a consumer home automation device I have (3+ diff ways).
2. The vendor doesn't only serve consumers, but also critical infra (announced partnerships include pharma manufacturing plants, airport logistics, hospitals, etc) across various locations in Europe and the US (not the same product line, ofc, so maybe moot).
3. Claude and Astra refuse to help me explore even my own device, so now I'm using Chinese hosted models for investigating.
4. The vendor (with a revenue of $0.75B) does not maintain any security disclosure practice whatsoever, nor does respond to emails.
So, to recap, I need to outsource vulnerability hunting of (potentially, I don't know) critical western infrastructure to Chinese models, because the western models refuse; and the vendor doesn't give a shit. Fun timeline.
PS: The vendor is Swiss.