Building autonomous AI agents. Notes from the team building Gravity. AI workflows, the future of recurring work, and what we learn along the way.
An agent that never fails does not exist. APIs time out, tools return garbage, a model invents a step that was never asked for, and a permission check blocks an action halfway through a task. The question that…
Read post →
A team workspace for AI agents is a shared space where everyone buys, runs, and governs the same agents under one set of rules, instead of each person quietly building their own. The short answer to "how do I create…
Read post →
Connecting an AI agent to a private API is what turns a chatbot into something that does real work. The agent stops describing what could happen and starts reading your inventory, creating tickets, or pulling a…
Read post →
A data pipeline moves data from where it lives to where it is useful, cleaning and checking it on the way. The hard part is rarely the happy path. Sources change a column name overnight, a feed sends half its usual…
Read post →
To audit an AI agent's decision history, you reconstruct a single run from its log: the input it received, the plan it formed, every tool it called with the arguments and results, the final output, who triggered it,…
Read post →
A travel advisor's real product is judgment: knowing which resort suits a honeymoon, which layover is worth the savings, which supplier answers the phone when a flight cancels. Yet most of an advisor's day disappears…
Read post →
Insurance broking is an admin-heavy trade. An independent broker juggles multiple carriers, chases documents that never arrive on time, and tracks renewal dates across hundreds of clients. The job is advice and…
Read post →
Event planning is a high-volume coordination job dressed up as a creative one. For every hour you spend on design or guest experience, you spend several more chasing vendors, updating spreadsheets, and answering the…
Read post →
A prompt is not documentation; it is the logic of the agent. Change a line and you change how the agent behaves, sometimes for the better, sometimes in ways you only notice three days later when results quietly…
Read post →
An agent that passes every test in the lab can still fail on its first real day. Test data is tidy; the real world is not. Going live is its own discipline, separate from proving the agent works on known cases, and…
Read post →
Some jobs are too big for one agent. Producing a competitor analysis means researching several companies, summarising each, and assembling a comparison; doing all of that in a single agent stretches it thin and makes…
Read post →
Most useful agent work is not a single action. Chasing an overdue invoice means pulling the ledger, finding what is overdue, drafting a message for each client, sending it, and logging the result. That is a…
Read post →
The fastest way to lose trust in an agent is to let it do something irreversible that it got wrong. The fix is not to supervise everything, which would defeat the point of automating, but to design the agent so a…
Read post →
Ask the same agent to do the same thing twice. If it does exactly the same thing both times, it is deterministic. If it might take a slightly different path the second time, it is probabilistic. That single property,…
Read post →
Every agent has to answer two different questions. First, what should I do to reach this goal. Second, how do I actually do it. The first is planning. The second is execution. They look like one smooth motion from…
Read post →
An AI agent can only think about what fits in front of it. That "in front of it" is the context window: the span of text the model reads before deciding its next move. For a quick question it never fills up. For a…
Read post →
A useful AI agent is rarely one giant instruction. Underneath, the good ones are assembled from smaller parts: a tool that reads a calendar, a tool that sends an email, a packaged routine for drafting a reply, a…
Read post →
An AI agent swarm is several agents working in parallel toward a shared goal, coordinating through a shared environment, messages, or a supervisor, with useful behavior emerging from their interaction rather than…
Read post →
The one-line distinction is this: a software robot replays a fixed script of clicks and keystrokes, while an AI agent reasons about a goal and decides its own steps at run time. The robot follows a recording. The…
Read post →
Here is the one-line distinction. A digital worker is a vendor's marketing package: a named, human-shaped persona, say an accounts-payable clerk, that bundles automation, often RPA bots, with some AI on top. An AI…
Read post →
An AI agent is not one thing. Under the surface, the agent follows a control structure that decides how it reasons, when it calls tools, whether it checks its own work, and how it splits a job into smaller jobs. That…
Read post →
Comments are the part of a WordPress site that never sleeps. They arrive overnight, on weekends, and in bursts after a post does well, and a chunk of them are spam, a smaller chunk are abuse, and the rest are the…
Read post →
On X (Twitter), the gap between a mention and a reply is where reputations are made or lost. A frustrated customer tweets, a journalist quotes you, a prospect asks how you compare to a rival, and a bot tags you to…
Read post →
By the time a customer clicks cancel, the decision was made weeks ago. The card that failed three times, the plan that got downgraded, the seats that stopped logging in, the support ticket that never quite got…
Read post →
A new product review lands on your Shopify store. Now the clock starts. A happy five-star note deserves a warm thank-you while the buyer still feels good. A two-star complaint about a torn seam needs a calm reply…
Read post →
Sourcing eats a recruiter's day. You read a role brief, search for people who fit it, read profile after profile, write a first message that does not sound like a template, wait for replies, chase the silent ones,…
Read post →
A spreadsheet is honest about almost everything except its own quality. The rows look fine until a date is stored as text, an email is missing the @ sign, a price has a stray currency symbol, or the same customer…
Read post →
Bank reconciliation is the chore that eats a bookkeeper's week and rewards none of the effort. Most lines match an invoice cleanly: same amount, the invoice number sitting right there in the bank narration. The work…
Read post →
A Webflow form is the easy part. The hard part is what happens in the thirty seconds after someone hits submit: is this a sales lead or a support question, has this person contacted you before, which channel should…
Read post →
The daily live standup is a tax on a distributed team. It interrupts deep work, forces people across timezones into one awkward slot, and most of what gets said is status that could have been read in thirty seconds.…
Read post →