In the late 1980s, Geary Rummler and Alan Brache gave operations people a phrase that never stopped being true: the greatest opportunities for performance improvement lie in the white space of the organization chart — the handoffs between the boxes, where work waits, context evaporates, and nobody is accountable because the gap belongs to no one.
Two generations of delivery practice took that seriously in the only way the technology allowed: by measuring the white space. Value stream mapping drew it. Value stream management instrumented it — flow time, flow efficiency, where a request sat idle, which queue ate the quarter. Mature delivery organizations today can tell you, with dashboards, that a change request spends most of its life waiting between functions. That was worth knowing. It was also, quietly, an admission: we could see the white space, price it, and manage around it — but we couldn't do anything in it. The gaps were where work slept, because no worker lived there.
That constraint just expired.
The first technology that works in the gap
Follow one change request through a typical enterprise today. A product owner writes it up. It waits for refinement. An analyst translates it for the architect. It waits. The architect shapes it for the build team. It waits for a sprint. QA receives it with a fraction of the original intent intact. Five handovers; at each one, knowledge is lost and time is spent re-establishing context that the previous box already had. The dashboards call this wait time. It is really translation time — the cost of moving intent between heads.
Agents change what the white space is made of. A capable agent doesn't sit in a box on the chart; it sits in the gap, holding context that used to evaporate at each handover. The delivery loop engineers know — inner loop at the keyboard, outer loop through CI and review, the outermost loop through users and telemetry — collapses toward a single closed loop: a working session where a domain expert states intent, an agent drafts the spec and the build, verification machinery grades it, and the expert judges the result while the context is still warm. Not tickets between departments. Iteration inside one loop, with a human in it.
That is the shift I mean by value stream engineering: we spent a decade managing the stream — measuring it, mapping it, tightening its queues. Now we can engineer in it — put working systems inside the gaps the map only priced.
What this does to the org chart
Here is the uncomfortable implication, and I'll state it plainly: the functional-box org chart, with managers whose main job is coordinating handovers between the boxes, is no longer a defensible design. Its whole justification was that translation between specialties was expensive and error-prone, so it needed dedicated coordination. When the translation cost collapses, the coordination layer is managing a problem that is dissolving under it.
What replaces it is not flatness for its own sake. The control objectives all survive — reviews, approvals, audits, segregation of duties. What changes is the enforcement: from meetings to mechanisms. A permission line an agent cannot cross does the job a sign-off meeting used to do, faster and with a cleaner audit trail. The gates stay; they become machinery with named human owners, rather than calendar events.
The honest caveat
The mechanics of this are provable — you can build the loop, wire the gates, and measure the flow in weeks. The human dynamics are the hard part, and anyone who tells you otherwise is selling something. People built careers in the boxes. The analyst whose value was translation, the manager whose value was coordination — they need a real place in the new design (there is one: contract owners, gate holders, graph architects), and getting them there is change management of the classic, unglamorous kind. Rummler and Brache would recognize every bit of it. The technology collapses the white space; only the organization can decide to live there.
Start where VSM already told you to: take your most-measured value stream, find its single most expensive gap, and run one working session in it — domain expert, capable agent, the same controls enforced by mechanism. Measure the flow before and after. The dashboard you already own will tell you whether the thesis holds in your house.
A version of this argument appeared in my LinkedIn article "From Value Stream Management to AI-Assisted Engineering" (2026).