FRAMEWORK CHOICE · CONSTRAINTS FIRST
Agent framework landscape: choose by constraints, not popularity
Score four agent framework archetypes for a workload, and see when popularity outvotes a requirement the workload cannot give up.
Event trace
Computed data
Final metrics
The four options are archetypes with invented 1–10 ratings, not reviews of named products; check any real framework against your own requirements. In soft mode each met requirement adds two points to the weighted sum. Licensing, cost, team skills and vendor support are not scored.
Popularity outvotes a requirement
The workload requires that data stays in your network, that the framework runs in-process on the JVM and that it offers a human approval step; requirements are only folded into the score. With weights of 2 for prototype speed, 2 for operability and 5 for popularity, the Hosted agent platform scores 83 and wins although it fails two requirements: VIOLATES CONSTRAINTS. The Python graph library scores 80, the low-code builder 71 and the JVM agent SDK, the only option that meets every requirement, 55.
Requirements as filters
Tick Apply requirements as hard filters before scoring. Three options are excluded and the JVM agent SDK is chosen with 49 points, the weighted score without bonuses: FITS CONSTRAINTS. A met requirement is worth only 2 points in soft mode, so no weighting rescues it: with popularity at 0 the hosted platform still wins, and with operability at 0 the Python library wins at 70.
Change the workload
Untick every requirement to model a public prototype: the hosted platform wins with 81 and fits, because there is nothing to violate. Add Requirement: works air-gapped to the default set and the hosted platform still wins at 83 in soft mode, now failing three requirements. With hard filters, air-gapped and approval required but the JVM requirement removed, the Python graph library fits and wins with 76.
Reading the table
Each option row shows the 1-10 ratings, which requirements it meets, its score and whether it was chosen, excluded, violates a requirement or fits. The board shows the weighted scores, coloured red for excluded options and amber for options that violate a requirement in soft mode.
Replay and limits
Any change recomputes the run and rewinds it; the Complete-run preview states the outcome at once. Play and Pause walk through the events, Next event jumps one event, Restart rewinds, Speed runs from 0.25x to 2x and Inspection time seeks from 0 to 60 seconds, where the outcome and final metrics appear. The four options are archetypes with invented ratings, not reviews of named products. Licensing, cost, team skills and vendor support are not scored; check any real framework against your own requirements.