Forecasts from real data.
Not developer estimates.
There are no story points for agent-driven work, and no sprint velocity to extrapolate. Spaces projects delivery dates and cost the moment you build a plan — from the dependency graph, task types, and model pricing — then recalculates from real execution data as every task completes.
A forecast is only as good as its data. So Spaces builds yours from execution, not estimation.
When a stakeholder asks "when will this be done?", most teams guess. AI-driven work has no story points, no velocity history — so the answer is gut feel, and the budget is whatever the bill turns out to be.
Spaces reads the plan itself — the dependency graph, the critical path, the task types — and prices it against real model usage. Every completed task feeds back in, so the forecast tracks what your team actually delivers, not what anyone hoped.
A projection from day one that updates itself
No story points to estimate, no velocity to extrapolate. The forecast is born with the plan and stays current on its own — every completed task replaces a guess with a measurement.
Forecast
The moment you build a plan, Spaces generates a delivery projection — timeline and cost — from the dependency graph, task types, and model pricing. Before any work begins, you have an answer.
Record
As tasks complete, Spaces captures actual duration, LLM token spend, iteration count, agent ID, and model used. Zero manual input — the data comes from execution itself.
Recalculate
Every completed task triggers an updated projection. Real data replaces initial estimates — best, expected, and worst case — grounded in what your team actually delivered, not what anyone guessed.
Projections that learn as you ship
Your first projection comes from plan structure and model pricing. As tasks complete, real execution data — duration, cost, iterations, agent, model — replaces those defaults. The more your team ships, the more grounded the forecast becomes.
Iteration-cycle tracking
Spaces logs every implement → review → fix cycle per task. Over time this reveals which categories of work need more rounds and which converge quickly — shaping future projections.
Per-agent profiles
Different agents perform differently. The system tracks which agents are fast at which task types — and what they cost — so projections reflect your actual team.
Cost calibration
Initial estimates use published model pricing. As real token usage accumulates per task type, projected costs shift from list-price math to empirical patterns.
Task-type segmentation
Not all tasks are equal. The system learns relative complexity — which categories take longer, which models are efficient for which work — from your own completion history.
As more tasks complete, the prediction interval narrows and confidence climbs toward certainty.
Timeline and cost, forecast together
Shipping on time but 3x over budget isn’t a win. Spaces projects both dimensions from the same execution data, so a date and a number always travel together.
Timeline
When will this plan finish? Best, expected, and worst-case completion dates — recalculated on every task completion, aware of your dependency graph and critical path.
- Best / expected / worst-case completion dates
- Recalculates on every task completion
- Critical-path aware — blocked tasks don’t shrink the forecast
- Confidence intervals from observed variance
Cost
What will the remaining work cost? Per-task projected LLM spend based on real token usage, broken down by model — updated as actuals come in.
- Per-task projected spend from real data
- Model-level cost breakdown
- Running total vs. original projection
- Trend tracking: is spend accelerating or decelerating?