Matthew Berman on Long-Running Coding Agents and Model Routing

Matthew Berman8m 55s
0 comments · 0 votesOpen discussionClose discussion
Sign in to join the discussion

    Video summary

    Matthew Berman demonstrates a spreadsheet-style web application produced during a coding-agent session that he says ran for several days before he stopped it. Sorting, formulas, validation and a pivot-table example illustrate useful progress. Matthew Berman also acknowledges rough edges and that the application covers only a subset of the original software's functionality.

    Matthew Berman describes the agent alternating between desktop inspection and implementation, then shows a game prototype from another extended session. These selected demonstrations illustrate a workflow involving tools and repeated work; they do not establish exhaustive correctness, complete compatibility or safe unattended operation on every task.

    Matthew Berman discusses choosing model sizes and reasoning settings for planning and implementation. His comparisons with other models are personal impressions at recording time, not a standardized independent ranking. The sponsored benchmark segment, current-price assertions, quota-burning encouragement and promotional links are excluded.

    Original YouTube thumbnailWatch on YouTube

    Share this page

    Matthew Berman against black beside the blue-and-white headline "PROTOTYPE IS NOT PARITY". Framed in blue with WWW.ARTIFICIAL-INTELLIGENCE.VIDEO, 9 July 2026 and duration 8m 55s.

    Matthew Berman shows prototypes from long-running coding-agent sessions, while the demonstrations remain partial implementations rather than proof of complete feature parity.