Cross-note synthesis

Building links between technical experts and reformers

How the published AI CoLab notes, taken together, speak to the Building links between technical experts and reformers evaluation theme. Public, general synthesis — safe to draw on without checking. It generalises the lessons; the sources below link back to the particular workshops (some members-gated).

56 insight notes so far. Not every workshop becomes one — recurring sessions and smaller gatherings usually don't. For some of the larger workshops a transcript is turned into an insight note so its lessons can be shared — with the room's consent, through time, and with people who weren't there.

Links form around translation artefacts, not introductions. In every note the bridge between technical experts and reformers is a concrete thing both sides can interrogate: tools co-designed on real, successful source material with the people who hold lived experience (Unlocking career potential); a working technical workflow made legible to policy, program and research staff (AI for Insight Notes); shared use cases discovered across organisations, with explicit prompts to follow up (Growing with AI); and a practitioner case study set beside a consulting framing, so reformers see an in-government example rather than a vendor pitch (AI for the Systems You Inherited); a shared fictional disaster the room built and then repaired together, which connected an international observer of government transformation with practitioners across policy, delivery, IT and academia (When the Algorithm Goes Rogue); a professional ethicist's theory turned into an exercisable process the people who own the problems could run themselves (Beyond the Principles); and a working proof-of-concept data platform that put AI engineers and data scientists in direct conversation with the custodians and heavy users of national datasets, with ideas flowing both ways — from pipeline architecture to what stakeholders actually need (Tourism data); and a live derivation of how the technology itself works — from regression to the attention formula — that let an industry AI researcher reason about model behaviour alongside policy, governance and delivery staff, with a chain-of-trust framing for passing assurance upward rather than handing the maths to the decision-maker (Beyond the Black Box); and a live demonstration of a production system connecting an AI agent to authoritative registries, which put private-sector legal-technology engineers in direct conversation with public-sector data custodians about pro-bono and public-good uses rather than a commercial pitch (Authoritative by Source); a live AI assessment participants could both undergo and steer, which put a system-builder in direct conversation with workforce-and-capability leaders, assurance staff and human-centred practitioners about the controls themselves (Getting Hands-On With High-Risk AI); and a deliberately experience-mixed foresight exercise that seated builders, strategists, designers and policy people at the same tables and modelled facilitation techniques they could take back to broker the same conversations themselves (Beyond the Hype); and a live demonstration of one agency's own production agents, offered alongside an explicit invitation for other agencies to reuse the underlying patterns — "the design travels, the build doesn't" — with a flagged buildathon and white paper as concrete next steps, so the artefact is a reusable in-government pattern rather than a product to buy (From Pilot to Production); and a role-based exercise that translated agent, memory, context and orchestration constructs into terms managers, policy staff and service owners could govern, using a concrete enterprise-transformation example to connect "how it works" to "how you lead it" (AI Productivity in the Agentic Era). The earliest sessions already worked this way: an audit framework used as a shared boundary object that oversight bodies and delivery teams could both hold (Auditing AI); a sovereign tool co-designed with the high-stakes practitioners who would rely on it (SCOTi Flow); and a modelling exercise in which distinct AI agents each stood for a different facet of a wicked problem, giving a mixed room a common object to reason over (AI for problems humans can't solve). The artefact is sometimes a method rather than a tool — but the move is identical: make the expertise into something both sides can pick up and use.

The clearest case makes the artefact a jointly authored one. In Human-AI teaming the model acted as a translator across disciplines — capturing technical language and reframing it in policy terms — so a co-authored draft strategy became the shared reference connecting anthropologists, mathematicians, lawyers and economists who would otherwise talk past each other. The same translate-into-a-shared-object move recurs where the subject is forbiddingly technical: a red-teaming exercise turned model behaviour into a policy-and-service problem the whole room could work (Government information in the age of AI), and structured walkthroughs let participants from engineering, defence, economics and policy reason together about compute strategy (AI and the geopolitics of compute) or interrogate contested economic models side by side (Exploring the economics of transformative AI).

The bridge increasingly reaches beyond the room to the affected. Where the earlier links connect builders to reformers, Beyond the Principles extends the obligation to the people a system acts on: it treats adoption as a joint study across governance, HR and research functions and argues affected communities should help shape a tool rather than be consulted after the fact — the same co-design instinct Unlocking career potential applied to those with lived experience of the problem.

The traffic is becoming two-way. The earlier notes mostly move technical detail toward reformers — explaining how the tools work so non-technical staff can ask sharper questions (AI for Insight Notes). AI for the Systems You Inherited argues the reverse flow is now mandatory too: as AI absorbs routine implementation, engineers' value shifts toward engaging business stakeholders directly. People and Culture Leadership names the role that holds the two ends together: people-and-culture leaders sitting at the table with technologists not to ask "but what about the people" from the sidelines but to translate between technical and human framings and steward the workforce side of the change. Vibe Coding for Curious Humans shows the same renegotiation at the smallest scale: a non-engineer's working prototype becomes the brief an engineer then hardens into production, so the reformer hands over a thing to build from rather than a document to interpret — loosening the engineer's gatekeeping role without retiring the depth that makes the prototype robust. The link-building obligation is landing on both sides at once.

Connection inside organisations is as scarce as connection between them. Growing with AI recorded colleagues from the same department meeting for the first time, and closed with a call for government and research bodies to set consistent frameworks so the sector adopts AI coherently rather than in silos. The workshops are functioning as connective infrastructure the org charts do not provide — which is itself evaluation evidence for the CoLab's convening role. AI and the changing nature of information work makes that role the object of study: a UNSW research programme takes the CoLab community as a case for how sense-making about AI travels back into agencies, and participants named neutral, cross-sector space — where agency identity and hierarchy can be set aside, failures shared candidly, and precedents set — as a structural enabler of adoption, since precedent is the most powerful artefact in a bureaucratic environment.

Sources

Last compiled 2026-07-03 from 56 published note(s).

Notes contributing to this theme:

Shared with the AI CoLab Alliance — see join.aicolab.org.
This note was synthesised from the workshop recording with AI assistance and human review. It captures general lessons only, is not attributed to individual participants, and was reviewed for sensitive content and shared with the presenters before publication.