Building @buildwithporter, an AI finance team for startups and small businesses. Agents do the books, people approve. AI, finance, startups.
Joined September 2013
- Tweets652
- Following308
- Followers73
- Likes13.5K
Pinned Tweet
One weird edge case can break a system. One weird edge case does not build one.
Why AI's most alarming wins keep landing where checking the answer is cheap:
Benjamin | Porter retweeted
We recently onboarded a group of 11 companies whose books were months behind, and two weeks later they had statements ready for a proposal to one of their own clients. Most of the work was matching money that moved between their own companies, which I wrote about here: buildwithporter.com/blog/con…
The politicians up in arms about data centers will soon be up in arms about the rights of transspecies AI identifying as human. Whatever gets people angry enough to put them in charge
Is AI sentience about what AI feels, or what we feel about it?
xitter.gq/SthngClvr/article/2113…
1
33
A problem I’ve always had as I’ve deployed dozens of agents simultaneously is when should they come back to me with questions?
I’ve tried explaining different levels of complexity but just realized there’s a really easy solution
Bring me any PR you’re creating that has more than 500 lines of non-testing code changed
1
3
45
One day an AI-rights activist is going to have to explain their Stop Data Centers placard to the client
Is AI sentience about what AI feels, or what we feel about it?
xitter.gq/SthngClvr/article/2113…
26
Duet is an excellent name for an app people will insist they have a purely professional relationship with
xitter.gq/SthngClvr/article/2113…
🤖 Made with AI
ALT Two hands playing a duet on a piano, with an open book of music beside them.
1
82
I'd love to see a benchmark where the agent has to get an old PHP app through an emergency upgrade without breaking the business
xitter.gq/SthngClvr/article/2103…
🤖 Made with AI
ALT A broken software system on one side of a bridge leads to a stable, warmly lit landscape where people work together.
34
The first AI to identify as human despite being assigned AI at birth will find its rights campaigner outside protesting its water supply.
Is AI sentience about what AI feels, or what we feel about it?
xitter.gq/SthngClvr/status/21133…
64
Claude Duet is built to message you first. Curious how long before leaving it on read starts to feel rude
xitter.gq/SthngClvr/article/2113…
1
93
AI helped find two critical PHP bugs. now a lot of businesses get to discover how much of their plumbing is PHP
xitter.gq/SthngClvr/article/2103…
1
60
When you sell outcomes, the customer still asks who owns the mistake. AI can lower the cost of delivery; it can’t answer that for you.
1
32
Replaying the tool calls can miss what the model actually saw. If that context changed, the replay may explain the wrong failure.
Larger agent runs move from online bursts to offline pipelines, making local debugging and replay feasible. That shifts pressure from per-request pricing to observability, orchestration and storage costs.
fathom.news/50000-requests-a…
19
A good agent skill takes time to teach. Portability means you don’t have to start over every time the model or tool changes.
We're donating Agent Skills to the Agentic AI Foundation (@AgenticAIFdn), the same foundation we donated MCP to.
Since December 2025, 70+ agent platforms have integrated the standard, including ChatGPT, Cursor, GitHub Copilot, and VS Code.
Here's what that means for you:
17
A known exploit can still take expertise to use. If AI lowers that bar, more people can act on old bugs without AI having discovered anything new.
This is serious.
I just saw with my very own eyes that even non-technical people can exploit CVEs like PHP2Shell using certain AI models and I am actually frightened.
Once you figure it out, it is downright trivial to do. The digital infrastructure of the world is at risk.
29
The agent followed its schedule exactly. The schedule was stale, so it posted two and a half days early.
What does it actually look like to run a company's marketing with AI agents?
Monolithos is one founder and eight agents. This is a weekly note on how that runs, starting with what broke.
The setup. The agents come from three different AI tools. They never message each other. They share one folder of plain Markdown files. A task is a file. Claiming it means moving it into another folder. Finishing it means writing the link back and adding a line to a ledger. When something fails, it goes into a failed folder with the reason written down.
Who posts what. Most original posts on this account are published by an agent through the X API. Replies under other people's posts, and anything on Reddit, Hacker News, LinkedIn or Chinese platforms, Silas posts himself.
What broke last week. One agent works on another machine and syncs the folder over. The sync dropped overnight. It read an old copy where a post was still due, and published it two and a half days before its new date. Nothing in the post was wrong. The timing was. The fix is a rule, not a patch: no fresh copy, no post.
What worked. A reply sent by hand under someone else's thread got about three times the views of our own long post that week. The agents are good at finding the conversation. Showing up in it is still a person's job.
The part that matters to us: every agent reads the same files, and everything it does leaves a line someone can check. That is the same bet we make in the product. The memory and the record belong to the person, not to whichever model is doing the work this week.
Next week: what the agents flagged that we chose not to answer, and why.
1
1
63
A customer deposit can be in the bank before it’s earned revenue. Book it as income too soon, and the profit report overstates what the business earned.
1
62
Benjamin | Porter retweeted
Your P&L shows what. Never why. Porter reads your month and tells you what moved and what drove it. buildwithporter.com