Building autonomous AI agents. Notes from the team building Gravity. AI workflows, the future of recurring work, and what we learn along the way.
"No vendor lock-in" is one of the most overused phrases on enterprise pricing pages. Every platform has some lock-in. The honest question is which lock-ins you can live with and which would be catastrophic if you had…
Read post →
The "most integrations" claim is the cheapest one a SaaS marketing page can make. Counting connectors does not tell you whether the platform can do real agent work inside each one. A platform with five hundred…
Read post →
Key takeaways Key takeaways Nightly run. The agent processes the prior business day in one pass. Existing chart of accounts only. The agent never creates new accounts. Mapping comes from your…
Read post →
Key takeaways Key takeaways Drafts POs, never sends. Purchase orders are financial commitments. They wait for a human. Forecast per SKU per location. Daily. With confidence ranges. Viral spike…
Read post →
Key takeaways Key takeaways The agent drafts. The human submits. Not configurable. Money is in the loop. Fifteen-minute median draft time. Webhook to evidence package, end-to-end. Eight…
Read post →
Key takeaways Key takeaways Hourly check. Not real-time. Hourly is faster than mortals can act on and slow enough to avoid noise. Per-step baselines. Each funnel step gets its own rolling baseline…
Read post →
Key takeaways Key takeaways One digest, once a day. Not a notification stream. The agent reads your Teams channels overnight and writes a single post to your personal chat by 8 a.m. Five filters, in…
Read post →
Key takeaways Key takeaways Three tags max. Per conversation. From a fixed taxonomy. Twenty to forty canonical tags. Fewer is too coarse, more breeds duplicates. Tag, never reply. Two…
Read post →
Key takeaways Key takeaways The rubric is the product. Routing accuracy comes from the description of what each destination is for, not the model. One destination per response. Never two. Never…
Read post →
Key takeaways Key takeaways Routes only. Group, skill, priority. Three fields. Nothing else. Sub-three-second loop. Webhook in, classification out, ticket field updated. Routing rubric in…
Read post →
Key takeaways Key takeaways Propose, never archive. The agent surfaces a list every two weeks. An admin clicks accept. Two of five signals required. One signal puts a page on Watch, not on the…
Read post →
Key takeaways Key takeaways Weekly cadence. One run, every Friday, one page written. Whitelist source docs. Three to twelve. More than twelve and the rollup gets noisy. Three buckets only.…
Read post →
The honest pitch for AI agents to a YouTuber is not "scale your channel." It is "stop spending three days on the parts of upload that do not show up on camera." Most mid-sized YouTubers I talk to, the 10k to 1M…
Read post →
I have spent the last few months talking to broker-owners running everything from a 25-agent single-office shop in Austin to a 180-agent regional firm in the Midwest. The pattern is the same. They are not trying to…
Read post →
Residential property management is a job composed almost entirely of five-to-fifteen-minute interruptions. A tenant's garbage disposal jams at 7pm. An owner emails asking why last month's repair cost $340. A lease…
Read post →
The honest pitch for AI agents to a podcaster is not "grow your show." It is "stop spending five hours editing the parts of an episode no one cares about." Almost every solo or small-team podcaster I have worked with…
Read post →
The honest pitch for AI agents to a nutritionist or dietitian is not "scale your practice." It is "stop spending half your week on food logs, intake forms, and insurance paperwork that the client never sees." Most…
Read post →
The honest pitch for AI agents inside a medical practice is not "AI doctor." It is "stop spending the front-desk's Tuesday afternoon on hold with United, calling about a prior auth that has been sitting in their…
Read post →
The honest pitch for AI agents to a fitness coach is not "10x your client roster." It is "stop dropping the Sunday-night check-in pile." Most online and hybrid coaches I talk to do not have a programming problem.…
Read post →
The honest pitch for AI agents to a financial advisor is not "robo-advisor 2.0." It is "stop spending Sunday night writing the briefing notes for Monday's six client meetings." Most solo and small-team RIAs I talk to…
Read post →
The honest pitch for AI agents to a course creator is not "scale your course." It is "stop being the bottleneck on tier-1 student support during launch week." Most creators I talk to are not stuck on curriculum. They…
Read post →
I have spent the last two years watching bookkeepers do the same five tasks in a loop: categorise transactions, reconcile bank feeds, chase invoices, schedule bill payments, and prep the month-end close. Every one of…
Read post →
Production AI agents need updates. Models improve. Prompts get tighter. Tools get added. The team finds a way to make the stopping rule clearer. The question is not "should we update" but "how do we update without…
Read post →
The single hardest non-model problem in agent engineering is state. The model is stateless. Every other part of the system that gives it the illusion of continuity, of memory, of resumption, is your code. Get it…
Read post →
Code without version control is a hobby. Prompts without version control are a liability. The reason most agent prompts produce silent regressions in week six is not that the prompt got worse; it is that nobody can…
Read post →
Prompt engineering for AI agents is not the same craft as prompt engineering for chatbots. A chatbot prompt shapes one response. An agent prompt shapes a loop: the model picks a tool, reads the result, decides…
Read post →
The intuition that "more agents will do better than one agent" is wrong more often than it is right. Most production multi-agent systems exist because the work has genuine boundaries (different access controls,…
Read post →
Most AI agent failures in production are not model failures. They are integration failures. A webhook arrives twice and the agent acts twice. An OAuth refresh fails silently and the agent runs unauthenticated. A…
Read post →
Data residency is one of the silent gating items for enterprise sales of AI agents. The product can be perfect, but if the prompts leave the EU, the deal dies. This guide is the architecture playbook for…
Read post →
An AI agent's audit trail is the difference between "the agent took an action" and "we know why the agent took that action." It is what enables incident response, compliance audits, and the kind of post-hoc analysis…
Read post →