← All dispatches

May 2026 · Design

How we'd build a learning platform today

A series on the design choices behind CoNoggin. The first piece is about the field a goal-authoring system has to be fluent in.

We started with purpose.

The primary goal of every L&D initiative is change — so we made change the unit of delivery. Every activity, every programme, every piece of content attaches to a goal. Every goal opens a loop that measurement closes. Goals and impact are how the platform is organised, and how you engage with it.

Which leads to the structurally hard question:

How do you structure a goal in a language that lets AI use existing research to design the change well?

The two-field shape most goal-tracking tools ship with — desired outcome and how we'll know— doesn't carry enough of the right kind of information for any serious methodology to land cleanly. To pick the right intervention, the system has to be fluent in the field. The rest of this post is the field, briefly mapped.

What the research tells us

There isn't one school of thought called how change happens. There are about ten, each answering a different question. They overlap, they argue with each other, and most goals worth setting need at least three of them taken seriously at the same time.

The next ten sections walk the field. Each one names what the cluster claims, the frameworks worth knowing, and where each cluster's tools succeed or fall down.

Is this the kind of problem with a right answer, or one we have to feel our way through?

Sense-making. The most consequential question to ask before anything else, and the one most goal-authoring tools never ask.

Wicked vs Tame problems(Rittel & Webber, 1973). Tame problems have stable definitions and a knowable answer. Wicked problems don't — they're contested, novel, and there's no test for done. Cynefin (Dave Snowden, 2007) sharpens this into five domains: Simple · Complicated · Complex · Chaotic · Confused. Each demands a different action. Best practice applies in Simple. Expert analysisapplies in Complicated. In Complex you can't plan — you probe, sense, respond, running small experiments and letting the next move emerge. VUCA (US Army War College, 1990s) and BANI (Cascio, 2020 — Brittle · Anxious · Non-linear · Incomprehensible) name the texture of the operating environment. Polarities(Barry Johnson) flag that some “problems” aren't problems but tensions to be held — autonomy and alignment, structure and flexibility — and solving them as problems makes them worse.

Works when: the problem is genuinely uncertain or contested, and the choice between plan it and experiment your way through it matters. Falls down when: the work is routine — classification becomes overhead. The hidden cost of skipping it: most AI-adoption goals are Complex but get authored as Complicated, which is why so many feel hollow on contact with reality.

How do groups of people change what they do?

Change management. The dominant L&D field, and the one most existing goal-authoring tools default to.

Lewin (1947) is the grandparent — Unfreeze · Change · Refreeze. Every model since is honouring or rebelling against him. Kotter's 8 Steps (1996) is still the dominant top-down narrative — urgency · coalition · vision · communicate · empower · wins · build · anchor — and Kotter himself updated it in 2014 (the 8 Accelerators) to run those phases concurrently through volunteer networks rather than top-down. ADKAR (Prosci, 1998) zooms in to the individual experience inside org change — Awareness · Desire · Knowledge · Ability · Reinforcement— and is currently the most common L&D framework on Earth. Bridges' Transition Model (1991) adds the layer most execution-focused models skip: the difference between change (the external event) and transition(the internal psychological adjustment). Recent work (Prosci's 2025 Eight Ways AI-Driven Change Is Different, McKinsey's Agentic Organisation, BCG's 10-20-70) flags that linear-phase models break under continuous-AI conditions where there's no stable end state.

Works when:the change is planned, the end state is stable, leadership has authority, and the people involved share an understanding of what's happening. Falls down when: the end state is moving (most AI adoption), authority is distributed, or the shift required is in identity, not just behaviour.

How does one person change a habit?

Behaviour science. Sharper, more rigorous than the change-management literature in many ways, and conspicuously absent from most L&D practice.

Stages of Change(Prochaska & DiClemente, 1983) is the foundation. People move through pre-contemplation · contemplation · preparation · action · maintenance, and the intervention you need depends on which stage they're in. Switch (the Heath brothers, 2010) — Direct the Rider · Motivate the Elephant · Shape the Path — explains why willpower is the wrong lever. Atomic Habits (James Clear, 2018) is the current pop default — cue · craving · response · reward · identity-based habits. Tiny Habits (B.J. Fogg, Stanford, 2019) gives the operational equation: Behaviour = Motivation × Ability × Prompt. If a behaviour isn't sticking, one of those three is the broken bit. COM-B and the Behaviour Change Wheel (Susan Michie, UCL, 2011) is the academic gold-standard — Capability · Opportunity · Motivation → Behaviour — used by the NHS and serious public-health practitioners. Nudge(Thaler & Sunstein, 2008) gives the choice-architecture vocabulary.

Works when: the change is a specific, observable behaviour and the conditions to perform it can be designed for. Falls down when: the change demands new mental models or identity shift, not behavioural adjustment — or the audience sits in pre-contemplation and the intervention assumes action.

How do grown-ups learn?

Pedagogy. The discipline of designing the experience itself.

Knowles' andragogy (1968) says adults learn when they know why, draw on their own experience, and the problem is real. Bloom's Taxonomy (1956 → 2001) is the depth ladder — remember · understand · apply · analyse · evaluate · create. Gagné's Nine Events (1965) underlies most LMS course structure whether the LMS knows it or not. Kolb's Experiential Learning Cycle (1984) is why do, then reflect, then try again outperforms learn theory, then apply. Process frameworks: ADDIE (military, 1970s) is the default waterfall; SAM (Allen, 2012) is its agile alternative; Action Mapping (Cathy Moore, 2008) is the discipline of stripping every course back to what people will do differently. The cognitive-science update — Make It Stick (Brown, Roediger, McDaniel, 2014) — is the most consequential thing in the cluster: retrieval practice (testing yourself, struggling to recall) beats re-exposure; spaced practice beats massed; interleaving beats blocking; desirable difficulty is a feature, not a bug. Ericsson's deliberate practice is the same insight at the expertise end.

Works when:there's a definable body of knowledge or skill and learners can practise it with feedback. Falls down when:the gap is judgement under uncertainty, or the “skill” is actually a developmental stage rather than a procedure.

How do you make a wish actually happen?

Goal-setting as its own discipline. Conspicuously absent from most change-management literature, despite being the very thing CoNoggin is helping people do.

Drucker's MBO (1954) is the grandparent. SMART (Doran, 1981) is the most-cited goal framework on Earth and frequently the most misused — achievable sandbags, specific becomes pseudo-precision. Hoshin Kanri (Toyota, 1960s) is the ancestor of OKRs: nested objectives across the org with explicit catchball — negotiating up and down. OKRs (Andy Grove → John Doerr → Google, mainstreamed 2018) pair qualitative Objectives with measurable Key Results on a quarterly cadence. 4DX / Wildly Important Goals (Covey & McChesney, 2012) adds lead measures — the observable behaviours that cause the result, distinct from the result itself. The most underused move in corporate goal-setting. BHAGs(Collins & Porras, 1994) are the long-horizon counterweight to over-quantification. WOOP (Gabriele Oettingen, 2014) — Wish · Outcome · Obstacle · Plan — is empirically more effective at producing actual behaviour change than SMART because it forces explicit obstacle anticipation. Theory of Change (foundations and international development) is the most rigorous causal-chain framework anywhere — long-term goal ← preconditions ← interventions ← assumptions, with verification checkpoints.

Works when: outcomes are measurable, the actor has reasonable control, and the horizon is bounded. Falls down when: outcomes are emergent, control is partial, or the horizon is open-ended — which is most strategic goals. Underused additions worth borrowing: lead measures (4DX), anticipated obstacle (WOOP), causal-chain explicitness (Theory of Change).

What's worth learning, vs. what's just nice to know?

Capability and performance support. The cluster that asks the most uncomfortable question in L&D: should this be a course at all?

70-20-10(Lombardo & Eichinger, 1996) — 70% experiential · 20% social · 10% formal — is usually quoted, rarely designed for. Mosher & Gottfredson's Five Moments of Need (2009) is the cleaner frame: training addresses new and more; performance support addresses apply, solve, change.Most things shouldn't be a course. They should be a job aid, a checklist, an embedded prompt at the moment of need. The capability vocabulary: T-shaped and π-shaped people, capability frameworks (Korn Ferry, SHL), the Dreyfus skill model (novice → expert). The frontier work is reorganising the function. Capability Academies(Josh Bersin, 2020) move L&D from courses to sustained, multi-modal, multi-year programmes tied to business outcomes. Skills-based organisation (Deloitte, 2022) replaces job descriptions with skills inventories — staff projects from a skill graph, reward skill acquisition. Working Out Loud (John Stepper, 2015) is the social layer of 70-20-10 with operational shape — small public commitments, peer circles, visible work products.

Works when: the gap is between current and target performance on a known task, and that gap is genuinely about knowledge or skill. Falls down when:the task itself is changing, or the “skill” is a posture rather than a procedure (judgement, taste, ethics) — performance support outperforms training in those zones.

How do people grow over a lifetime?

Adult development. The slow layer underneath everything else.

Self-Determination Theory(Deci & Ryan, 1985) is the foundation: people sustain change when three innate needs are met — autonomy · competence · relatedness.The most empirically grounded thing in motivation psychology, and the quiet reason ADKAR's Reinforcement phase so often fails. Carol Dweck's Growth Mindset (2006), Amy Edmondson's Psychological Safety(1999, mainstreamed via Google's Project Aristotle in 2015), and the GROW · CLEAR · ICF coaching frameworks structure most executive coaching. The frontier: Robert Kegan's adult development theory — adults move through Socialised → Self-Authoring → Self-Transformingstages of cognitive complexity. Most modern jobs ask Self-Authoring complexity from people in Socialised stages, which looks like a skill gap and isn't. Theory U(Otto Scharmer, MIT) argues most change models can't access deeperchange because they don't include a phase of letting go of the current operating logic.

Works when:the challenge is genuinely developmental — current cognitive complexity can't hold the demand of the role. Falls down when:the issue is mistakenly diagnosed as developmental — when it's actually a skill gap, an environment problem, or a motivation problem, the developmental frame patronises and stalls.

How do offices arrange themselves to keep getting better?

Organisational forms. Not what's the goal but what kind of organisation makes goals like this thrive?

Peter Senge's Five Disciplines (1990) and Wenger's Communities of Practice (1998) are the canon. The familiar present: the Spotify model (chapters · guilds · squads · tribes — adopted everywhere, then quietly distanced from by Spotify itself), Holacracy (Brian Robertson, 2007), Sociocracy 3.0. The frontier is the proof that radical decentralisation works at scale. Buurtzorg (Dutch home-care, 2006) runs ~14,000 nurses in self-managing teams with no managers, and outperforms hierarchical competitors on quality and cost. Morning Star (a US tomato processor) runs without managers via Colleague Letters of Understanding — peer-negotiated commitments. Haier's RenDanHeYi (China, 2005) restructured ~80,000 employees as ~4,000 microenterprises with internal markets. GitLab runs handbook-first— a public, version-controlled, ~2,000-page operating system, decisions written before they're enacted.

Works when: the org is willing to redesign structure, not just retrain people. Falls down when: leadership wants behaviour change without authority change — the structural prerequisites are absent.

How do you tell if a change actually happened?

Evaluation. The cluster that closes the loop.

Kirkpatrick's Four Levels (1959) is universally taught — Reaction · Learning · Behaviour · Results— partly because it's clean, partly because nobody's improved on it without sacrificing usability. Phillips ROI (1997) bolts on a Level 5 for ROI, rarely justified outside compliance contexts. The more usable popular tools: Brinkerhoff's Success Case Method(find what worked, find what didn't, learn from the contrast) and Most Significant Change (stakeholder-narrated stories, peer-selected — good for the Complex-domain goals where numbers mislead). The frontier: Outcome Mapping (IDRC, 2001) measures progress markers on the behaviours of boundary partners — explicit about influence vs control. Reverse Kirkpatrick (Brinkerhoff and others) is the move that helps most: start with the desired Level 4 result and design backwards. Most goals are designed forwards. Reversing the design dramatically improves quality.

Works when: the change is observable, attributable, and sufficiently downstream of the intervention. Falls down when: outcomes are confounded by external factors, or the change is in things hard to measure cleanly — culture, judgement, posture.

What do innovative orgs actually do?

Last, and orthogonal. Not a school of thought — operating practices of organisations defining the frontier. Many are implementations of the clusters above; the practice itself is what others copy.

• Pixar's Braintrust. Peer-review meeting for hard creative problems. No authority gradient. Candour as duty.
• Amazon's working-backward + 6-pagers. Every initiative starts with a press release plus FAQ. Decisions in narrative prose, not slides.
• Bridgewater's Principles + dot ratings. Codified operating principles, real-time peer feedback on every decision.
• Google's quarterly OKRs and blameless postmortems. Public objectives at every level; failure made discussable.
• Microsoft's growth-mindset rebuild. Cultural pivot under Satya Nadella from know-it-all to learn-it-all.
• Stripe's writing culture. Decisions written and archived. Meetings start with written briefs.
• Linear's “How we work.” Documented operating principles, public.
• Basecamp's Shape Up. Six-week cycles with shaped (not estimated) work; appetite over deadline.
• Continuous Discovery (Teresa Torres). Weekly customer touchpoints, opportunity-solution trees, small experiments in parallel with delivery.
• Buurtzorg, Morning Star, Haier. Already named above. The proof points for self-management at scale.

The pattern across them: goals as narrative, documented, queryable, peer-reviewable artefacts. Not as form-fills. Most learning and goal-tracking software optimises for the form-fill model and produces dead goals.

What's the currency across the field?

Across all ten clusters, a small number of structural inputs show up over and over. Different methodologies use different vocabularies, but underneath they're often asking for the same units of information. The convergence is striking enough to be worth naming.

Six that appear almost everywhere:

• The change — the state of the world the goal is moving toward.
• The behaviour shift— what people will do differently. Distinct from the change. Action Mapping makes it primary; ADKAR's Ability + Reinforcement depends on it; behaviour science treats it as the unit of analysis.
• The audience starting state— where the people involved are now. ADKAR's Awareness + Desire, Stages of Change, Dreyfus skill model — all need this.
• The urgency / why now — Kotter's first step; Fogg's Prompt; the strategic context every methodology hangs on.
• The evidence — how we'll know — Kirkpatrick, Key Results, Theory of Change. Universal.
• The time horizon — every methodology has cadence, even when implicit.

Three that are usually missed and that quietly do most of the work:

• The problem type — the Cynefin classifier. Is this Complicated (analyse-then-prescribe) or Complex (probe-sense-respond)? Choosing a Complicated-domain methodology for a Complex-domain problem is the most common authoring failure in the field.
• The lead measure — the observable behaviour that causes the result, distinct from the lagging result itself (4DX). Most goal-tracking tools conflate the two and lose the predictive signal.
• The anticipated obstacle— what's likely to get in the way (WOOP). Empirically, the single most effective addition to goal-setting. Almost no system asks for it.

These nine are the field's working vocabulary. Anyone designing interventions across more than one cluster — and most L&D practice does — ends up reasoning in roughly these terms whether they name them or not.

What stands out

A few patterns across the ten clusters are worth flagging.

The clusters argue with each other in ways that are productive. ADKAR's individual-level lens, Kotter's org-level lens, Cynefin's what kind of problem first. No single cluster carries the whole load. Treating one as the methodology is a category error the field has been making for decades.

The most consequential question is the least asked. What kind of problem is this?— the sense-making cluster — determines whether any of the other methodologies will land. A change-management plan applied to a Complex-domain problem isn't a bad plan; it's the wrong kind of plan. Most goal-authoring software doesn't think to ask.

Frontier organisations don't author goals as forms.They author as narrative — peer-reviewed, documented, queryable, downstream of recurring contact with reality. The field's structured vocabulary lives behind the scenes; the surface stays human. Almost all goal-tracking software optimises for the form-fill model and produces dead goals.

Next dispatch: deeper into sense-making — what kind of problem is this, and why every other methodology depends on the answer.

CoNoggin is built by Alt Shift Lab. We're in pilot with one client and opening to a small group later this year. Join the waitlist →