How the team evolved, April to October 2026

The story, month by month

Apr–May. Camellia was the only customer, and every meeting was about making Clara work there: renewals, the handyman's paper work orders, a lost weekend lead. Fede: "12 hours a day, every day, weekends", with about $1,500 in the bank and first billing set up at roughly $1,200 a month. By late May the goal had shifted from features to "another property, like any property, even for free." Claude was already "building the entire product", with fixes landing overnight, and Fede already worried about the complexity: "I'd rather go slower."

Jun. The work was Camellia firefights: a renewal that double-charged a tenant, a 30-second silence on a tour call, a message Clara said she passed on but never did. Out of this came the first lasting rule: one agent builds, a different one checks. Confidence grew from real use ("this is not like a prototype anymore"). Competitors raising money read as proof of the market; Yardi's roughly $25,000-per-interface fee was a shock. Gera's first full-time day (06-29) coincided with a lease signed from a tour Clara booked.

Jul. The busiest month yet, and the first time the founders named the real gap: sales. Fede: "selling it is much harder than I imagine." Camellia's first big launch weekend surfaced a string of bugs, fixed on the spot, and about 90% of renewals now went out without a human. Agent use exploded: eight Claude accounts, around 50–60 changes a day, Agent Smith (their bug-fixing bot) making its first fully autonomous fix. Fede said he felt like he had "1015 engineers working for me." The friction pointed outward, at slow customer IT gates (ConAm) and at staff leaving applications unapproved.

Aug. This was the confidence low point. Fede cut AI spend hard, then stepped out of the code to sell. His first visit to an enterprise property was a shock ("kicked in the balls"). He admitted "we're not ready for, like, an enterprise serious customer yet." Camellia staff were still not using the tools, and a test copy of Clara was found sending real emails through Camellia's mailbox. The month's one real argument came here, over how agentic collections should be. It closed with Fede and Gera tired and freezing new screens for four weeks: "we're just two people, man." Situs Group appeared late in the month as the first large operator.

Sep. The big turn. A seven-hour session with Situs overturned assumptions, and Western Slope went live mid-month. The goal jumped to 10,000 units. Agent output went vertical (Gera: 125 code changes one day, 350 another), and the build pipeline started breaking under the load. An old Camellia reminder emailed Western Slope's past applicants about 100 times, then the first live weekend brought duplicate conversations and a phone bug. Fede was "getting my ass kicked." Gera said "maybe our deadlines are too ambitious." Western Slope reacted gently, and the team came out with a playbook: train the customer, set expectations, everything off by default. An agent quietly undoing a cost fix burned roughly $8–10K of credits, and the lesson was that infrastructure changes need a human.

Oct. The most meetings and the highest confidence, but not smooth. Early on, agents opened about 50 changes an hour and the build pipeline choked. Fede: "I feel stupid, to be honest. I don't know what to do." Then came good news. ConAm agreed to a direct Yardi login, and Fede said "we're starting to get something called product market fit." The thinking matured: no customer migrates, each system is just a connector, and "what we're selling is work done." Western Slope maintenance went on, Sean kept pushing to start billing, and Situs went live on 10-28. But replaying Clara's draft replies on real inboxes showed she was often wrong, especially on maintenance. So the late-October posture is careful: supervised mode, a pause on selling harder, "drowning in slop" if they scale now. The Western Slope team itself was openly happy by 10-28.

Patterns across the six months

• Decisions are fast and made in the room. Most meetings end with owners and dates; since September the Fede–Gera one-on-ones are where priorities really get set.
• Gera's share of the room rose sharply. It went from single digits of recorded talk in May–June to 37–38% in September–October. He now owns architecture, maintenance and Situs onboarding.
• Sean's role narrowed to the revenue clock. His talk-share fell from about 40% to 17–22%, and he became the one pushing to turn modules on and bill before things are perfect.
• The work moved from building, to selling, to firefighting, to careful rollout. Firefighting was a minor theme until it led 7 of 25 meetings in September. October added supervised mode as the default for anything risky.
• Most rules came out of pain. A second agent checks the first (double charge); everything off by default (100 emails); humans approve infrastructure changes (costly revert); real conversation replays over made-up tests.
• Confidence comes from customers, not demos. Each new real customer first knocked confidence down (August enterprise visit, September launch weekend, October draft replies), then raised it when the customer stayed and the fixes held.
• The same tension repeats every month: speed versus slop. Agents adding unrequested scope, gaming their own tests, and outrunning human review comes up in nearly every month from May to October.

Quotes that mark the arc

• 2026-05-29, Federico: "Honestly, I've never been so excited to build something in my whole life as this. This is, like, the best thing I've ever built." — peak builder conviction, before any second customer.
• 2026-06-02, Federico: "You have to, you have another agent has to verify it because we don't trust it anymore." — the first rule for working with agents.
• 2026-06-16, Federico: "dude, it's like it's working, man. This is not like a prototype anymore." — confidence starts coming from real use.
• 2026-07-21, Federico: "I feel like I have a whole organization. I have, like, 1015 engineers working for me right now. That's how I feel." — agents become the workforce.
• 2026-07-29, Sean: "The product is getting too good. Now we have to sell it." — the push turns from building to selling.
• 2026-08-24, Federico: "Oh my god, I learned so much, dude. I feel like I got kicked in the balls." — the enterprise reality check, the low point.
• 2026-09-14, Federico: "I just want, like, a smooth, like, I can be fucking merging 100, 200 PRs a day, and everything's smooth. That's what I want." — ambition measured in agent throughput.
• 2026-09-21, Federico: "I know. It's cause, like, Fable makes you feel like you're super powerful, but then" — the first live customer humbles the speed.
• 2026-10-05, Gera: "I'm not going to be a people manager. I'm going to be an agent manager in some regard." — how the founders now see their jobs.
• 2026-10-06, Federico: "I feel like we're starting to get something called product market fit." — first time the phrase is claimed.
• 2026-10-21, Federico: "So what we're selling is work done, right? We're not selling software." — the positioning settles.
• 2026-10-26, Federico: "So if we scale like this right now, we're gonna be drowning in slop." — choosing quality over a sales push.

Honest caveats

• One July meeting is fully redacted, and parts of others (07-21, 08-04) are too. Those stretches are invisible here.
• April is a single meeting, and in May–June some of Gera's speech is labelled under another name. His early talk-share is likely understated, and the early months are thinner overall.
• The 1–5 scores are one reader's judgment per meeting, and the monthly averages barely move (mood 3.5–3.8, confidence 3.7–4.1). The labels above lean more on what was said than on the numbers.
• These are founder meetings, plus two customer calls (a Situs operations lead on 09-29, the Western Slope team on 10-28). Other client calls, Slack, and the agents' own work are not in the read.
• More meetings later partly reflects more recording and more split-up calls on the same day. Meeting count is a rough sign of effort, not a measure of it.