Most teams that say they use Kanban have a board. Columns for To Do, In Progress, Review and Done, cards moving left to right, maybe a swimlane for urgent work. What most of them don’t have is a work in progress limit on any column. Without one, the board is a status display. It shows the work. It doesn’t change how the work flows. A WIP limit is a cap on how many items can sit in a column at once.

A coding agent can open a pull request in minutes. The reviewer still has to decide whether the change solves the right problem, whether it is safe to operate and who will maintain it. A clean diff and a green test run help, but neither answers all three questions. Consider a request to add a retry to a payment client. The agent may produce tidy code, add a unit test and describe the change accurately.

A2 is done, and HyperCrux 0.2.0 is out. The new release setup published it automatically once the tests passed, without you clicking anything. What 0.2.0 adds: hypercrux export writes a whole database as plain text lines, and hypercrux import reads them into a new file. Going in and out keeps every value exactly, and exporting the copy gives the same bytes as the original. The same export and import are available in Go.

Teams building connected products face a design choice that kills agility: either the device depends on the server for everything, or you build a converter that keeps two data formats in sync. The device has readings and settings in its own format. The server wants those same values as SQL rows. Somebody writes a converter. Then somebody has to keep that converter alive and correct for ten years. The converter approach is why IoT teams move slowly.

Teams shipping features fast need infrastructure that gets out of the way. The reverse proxy is supposed to be boring: TLS termination, serve static files, route traffic to the app, maybe some health checks. In practice, the config grows one regex at a time, until routing decisions become cargo cult superstition. Someone changed a block six months ago and nobody knows what would break if you moved it. BareProxy is built on the insight that you can reason about routing only if you remove the things that are hard to reason about.

Teams that ship features fast need dashboards that answer questions instantly. The usual pattern is to recalculate answers on every refresh, or to ship all your data to a hosted analytics service and wait for it to ingest. Both approaches have the same problem: the dashboard either waits for the query, or you pay for scale you don’t need most of the time. In either case, answering a question about your product becomes a friction point.

Every developer has lived the setup nightmare. You clone the repo, follow the README, get halfway through the steps, and something breaks. You Google the error, find three different solutions, apply the one that almost works, and spend two hours on an environment that everyone else on the team got right. Meanwhile, the new developer at the desk next to you is still stuck on step one. The setup is so fragile that nobody wants to change it, and it drifts from the reality of how the tests actually run.

Security gates at the end of a sprint kill velocity. A coding agent can cut feature work time in half, but security teams that don’t understand what the agent does will block it. The agent reads code from your repo, issues from your tracker, snippets from web pages, and it writes back pull requests. One malicious instruction planted in that text can make the agent send a deploy token somewhere it shouldn’t.

A client asks for a fixed price. The vendor writes a spec, quotes a number, and both sides sign it feeling like they’ve eliminated risk. Six weeks in, the client wants a feature that wasn’t in the spec, the vendor calls it a change order, and the relationship that started as a handshake is now a negotiation over what “originally scoped” meant. That’s not a fixed price project gone wrong. That’s what most fixed price projects actually look like once real requirements meet a document written before anyone had built anything.

Cloudflare is promoting GPT-5.6 Sol through AI Gateway at what it calls half price, running through September 18, 2026. No promo code, no application. Point your requests at the gpt-5.6-sol model and the rate applies automatically. The promo rates are $2 per million input tokens, $10 per million output tokens, and $0.25 per million cache read tokens. Cloudflare gives the regular rates as $4, $20, and $0.50. Those numbers are worth holding up against the source.