Every July 1 we run the same 40-minute exercise before touching a single goal: count the hours the quarter actually contains. Not revenue targets, not project lists. Hours. Every quarter that has ever gone sideways on us went sideways because the plan assumed hours that never existed.
Agents make this discipline more necessary, not less. They multiply what an hour produces, but they also bill you a maintenance tax that most people forget to book.
Here is the full Q3 math, with our numbers left in.
Start with the real number: 330 hours
Q3 is 13 weeks on the calendar and roughly 330 focused hours in practice. Subtract two full weeks for vacation, a holiday, and the sick days that will happen whether you plan them or not, leaving eleven working weeks. Then use a real ceiling of 30 focused hours a week, not the nominal 40, because email, admin, and being a human absorb the rest. Eleven weeks times 30 hours is 330.
Resist the urge to negotiate that number upward. The 40-hour version of the plan is the one that fails in August and takes your reputation for reliability with it.
330 hours sounds small. That is the point. Scarcity you acknowledge in July is scarcity you can design around.
Write the number where you can see it. Ours is on the whiteboard: 330. Every commitment for the quarter gets tested against it before anything gets a yes.
Price the agent tax before the agent dividend
Every production agent costs 45 to 60 minutes a week of human attention: reviewing diffs, repairing failures, updating memory, adjusting prompts when a model version shifts. Our seven agents cost about six hours a week, which is 65 hours across the quarter, a fifth of total capacity. Book that tax as a fixed cost before you allocate anything else, because it gets paid whether or not you planned for it.
The dividend is why you pay it. Those same seven agents produced work in Q2 that would have taken roughly 340 hours by hand: drafts, audits, reporting, follow-ups, site changes.
Net position: spend 65 hours, get back 340. But the tax is due weekly and the dividend arrives unevenly, so cash-flow the hours like money.
An agent that costs more upkeep than it returns gets retired at the quarterly review. We retired two in April. No sentiment.
Allocate by ratio, not by mood
Fixed ratios beat weekly judgment calls. Ours for Q3: 55% client delivery, 20% sales and marketing, 15% engine and systems work including the agent tax, 10% slack. On 330 hours that is 181 delivery hours, 66 for sales, 50 for systems, and 33 unassigned. The ratio is the strategy. Deciding it once in July means you never re-litigate it on a busy Tuesday when a client is loud.
The 20% sales block is the one solopreneurs cut first and regret by October, because pipeline sown in July closes in October. Cut it and Q4 starves.
The ratio also gives you a clean way to say no. A request that does not fit inside its block does not fit inside the quarter.
Different businesses need different splits. What matters is that the split exists in writing before the quarter starts spending you.
Cap client count with a formula
Your client cap is weekly delivery hours divided by honest per-client weekly cost. For us: 30 focused hours times 55% is 16.5 delivery hours a week. A standard retainer truly costs about 5.5 hours a week once you count calls, context switching, and revision rounds, not just production. So the cap is three retainers. Signing the fourth without changing the math means unpaid overtime dressed up as growth.
Most people lie to themselves in the per-client number. Track one honest week and include every Slack reply. The real figure is usually 30% higher than the guess.
Agents lower the per-client cost over time. Ours took a retainer from about eight hours to 5.5 over three quarters, which is how the cap went from two clients to three.
Project work fits the same formula, just lumpier. Convert each proposal into estimated hours per week across its run and test it against the delivery block before you quote a date.
The cap moves when the measured number moves. Never before.
Leave 10% slack or the plan is fiction
Slack is not laziness, it is load-bearing. The 33 unassigned hours absorb what the quarter will definitely throw: a client emergency, a model deprecation that breaks two workflows, an opportunity worth grabbing on short notice, a week where you are simply slow. Plans allocated to 100% shatter on first contact with any of these. Plans with 10% slack bend and hold.
Last quarter our slack went to an unplanned client rescue, six hours, and a migration when an API we depend on changed auth schemes, nine hours. Neither was on any list in April.
If the slack goes unused, it becomes the best kind of surplus: deep work on whatever compounds, usually the engine.
Unspent slack is a bonus. Missing slack is a debt with interest.
Run the recount in week seven
In week seven, mid-August for Q3, recount everything in 30 minutes: hours actually spent per block, agent tax actually paid, honest per-client cost, slack remaining. Compare against the July plan and correct while correction is cheap. A plan running 10% hot in August is fixable with one conversation. Discovered in late September, it is a missed commitment and an apology.
Our Q2 recount caught the delivery block running 14% over, one client quietly scope-creeping. One renegotiation later the quarter finished on plan.
The recount is also when agent metrics get honest review: which ones earned their tax, which are drifting toward retirement.
Thirty minutes in week seven buys the whole quarter's credibility. Put it on the calendar today, while July is still feeling generous.
What to do when the math says no
When demand exceeds 330 hours, you have exactly three honest moves: raise prices so fewer clients produce the same revenue, cut scope so each client costs fewer hours, or automate another slice of delivery so the per-client cost drops. Working more hours is the fourth option everyone takes and the only one that compounds against you. The math saying no is the system working.
We take the price path most often. A 20% raise at renewal, with the capacity reasoning stated plainly, has never yet lost us a client we wanted to keep.
The automation path is slower but permanent. Every function moved to the engine lowers the per-client cost for every future client.
Capacity is the one number that does not respond to optimism. Plan the hours, and the revenue follows the plan.