Building autonomous AI agents. Notes from the team building Gravity. AI workflows, the future of recurring work, and what we learn along the way.
A Linear sprint summary agent runs at the end of a cycle and produces a written record of what shipped, what slid, what the team added mid-cycle, and what got cut. It replaces the manual end-of-cycle write-up that…
Read post →
An Intercom auto-responder is a customer-facing surface, which means it has to do two contradictory things well. It has to respond quickly enough that customers feel heard, and it has to be wrong rarely enough that…
Read post →
A HubSpot lead scoring agent reads the data HubSpot already holds and produces a single score and a one-paragraph rationale for every contact. The SDR opens the contact, sees the score and the top three contributing…
Read post →
Most teams ignore Google Analytics until something is on fire. By the time something is on fire, the analytics signal has been visible for two or three weeks and nobody looked. A weekly summary agent removes the…
Read post →
An expense-categorisation agent does the same job a bookkeeper does in QuickBooks or Xero, but for every transaction the company posts, not just the ones that get to month-end. It reads the corporate card feed, reads…
Read post →
A Calendly follow-up agent listens for booking events, treats each invitee as a small workflow, and runs the sequence from confirmation to either a held meeting or a no-show recovery. It is not a drip campaign. It is…
Read post →
Asana inboxes accumulate. A user on a healthy team sees 60 to 120 inbox notifications a day: comments, mentions, status updates, task assignments, automation pings, due-date reminders. Less than 10% of these actually…
Read post →
A "watch list" agent is the simplest, most useful agent most people never bother to set up. It polls a small number of listings on your behalf, applies criteria you specify once, and alerts you when something…
Read post →
A weekly newsletter is the most resilient distribution channel a founder has. Algorithms change; inboxes do not. The cost is the time you spend assembling the issue. An AI agent can pull that cost down without making…
Read post →
The pitch for a meal-planning agent is simple: 30 minutes of weekly menu work, gone. The trick is that meal planning is bound by hard physical constraints (allergens, what is in the pantry) and soft preferences…
Read post →
Failed payments are the silent revenue tax of subscription businesses. A meaningful share of what gets called "churn" is actually involuntary: a card expired, a soft decline never recovered, an insufficient-funds…
Read post →
Slack channels accumulate three kinds of messages: the ones that need a fast reply, the ones that need an answer eventually, and the ones that need nothing. A triage agent's job is to make the first two visible and…
Read post →
Abandoned carts are the largest single conversion-recoverable signal a Shopify store has. Shopify's built-in flow recovers a meaningful share of them with a single email; an agent layer recovers a larger share with…
Read post →
The fastest way to ruin your LinkedIn presence is to point an AI agent at a generic prompt and tell it to post five times a week. The fastest way to keep your LinkedIn presence is to point an AI agent at a real…
Read post →
Late invoices are the cash-flow tax that small businesses pay every month. The work of chasing them is unloved, slightly awkward, and almost always overdue. An AI agent can carry the load, but only if it does the…
Read post →
Grocery reorder is one of the few agent use cases where the value is obvious and the failure modes are obvious. The value is the 30 minutes a week you do not spend rebuilding the same basket. The failure mode is a…
Read post →
Open pull requests pile up faster than maintainers can route them. Most of that delay is mechanical (which area is this, who owns it, which reviewer is least swamped) and machine-friendly. A PR triage agent does the…
Read post →
The single-agent vs multi-agent debate is one of the more confused conversations in AI agent design, partly because both sides frame the question as architectural when it is really economic. The technical question is…
Read post →
A single agent that calls tools is sometimes called multi-agent by marketing teams. It is not. A real multi-agent system has two or more agents with separate prompts, separate tool sets, and a protocol for talking to…
Read post →
"Training an agent on our docs" usually means fine-tuning in marketing copy and retrieval-augmented generation in code. The marketing version sounds fancier; the code version actually works for what most companies…
Read post →
Rate limits are the boring failure mode that takes out demos when they go to production. The agent runs fine on five test requests, then 500 requests arrive in a minute and a third silently disappear. This guide is…
Read post →
The first agent you ship has three tools. The third one has thirty. Somewhere between three and thirty, reliability cliffs and you cannot tell why. The cliff is real, the causes are predictable, and the mitigations…
Read post →
Most teams connect their first agent to Slack in 30 minutes using a personal token, then spend three weeks unpicking the consequences. This guide is the safe path from start: a bot app, narrow scopes, allow-listed…
Read post →
An agent that can put events on your calendar saves an hour a week. An agent that can put events on the wrong calendar costs you a customer. The difference is two layers of allow-listing, an explicit timezone…
Read post →
A human approval step is the cheapest insurance policy in agent operations and the most over-applied governance pattern. Done well it catches the actions that should never happen and stays out of the way for…
Read post →
"Is this an agent or just an LLM?" is the most useful question a buyer can ask in 2026, and the one most vendors avoid answering directly. The distinction is not pedantic. An LLM is a component; an agent is a system…
Read post →
The first time an agent does the wrong thing in production is the day a trust model becomes a budget line. Every team eventually writes one. The question is whether you write it before the incident or after. This…
Read post →
Safety for AI agents is structurally different from safety for chatbots. A chatbot that says something inappropriate creates a screenshot. An agent that does something inappropriate creates an incident: an email…
Read post →
The first time someone writes an agent prompt the way they write an LLM prompt, the agent breaks within the first hour of running. Not because the prompt is wrong in a literal sense; it is just shaped for the wrong…
Read post →
Eight months into 2026 and the agent market has visibly matured in three ways the 2025 forecasts mostly missed: pricing structures are converging, buyer procurement bars have hardened, and the marketing language of…
Read post →