Starting Where You Are

The framework’s earlier chapters describe what Trellis is. This chapter describes how it lands.

Most organizations encountering Trellis are not at zero. They have AI experiments running somewhere — in marketing, in customer service, in finance, in scattered places. They have governance postures of varying maturity. Some have AI principles documents written before any projects ran; some have nothing. They have an L&D function that may or may not already be tasked with AI fluency. They have leadership teams with mixed views about AI’s strategic relevance. They have a strategy document of varying quality and varying age.

Trellis does not start with a blank slate, because there are no blank slates. The framework is not an installation that gets switched on. It is an orientation that meets the organization where it is, makes sense of what is already there, and gives leaders a way to start moving toward what should be there next.

This chapter is about that meeting. It addresses the realities of starting, the conditions that have to be in place for the framework to operate, the recurring failure modes worth avoiding, and the signals that tell you it is working.

Translating where you are

Before the framework can help, the organization has to translate its current state into Trellis vocabulary. The translation is mostly diagnostic. The questions worth asking are these.

Do you have a working strategic loop, or a stale strategy document? A working loop is a current articulation of the four anchor questions that the leadership team can actually evaluate AI candidates against. A stale document is something written years ago that nobody disputes but nobody uses to make decisions. Most organizations have something in between. The relevant question is whether the strategy is alive enough to do the gate work in Chapter 9 — if the answer is no, the first move is to revisit the strategic loop before starting projects, not after.

Do you have AI projects running, or scattered experiments without canvases? Most organizations have a handful of AI projects in some stage. The diagnostic question is whether they are organized — running through something like a canvas, with named hypotheses about readiness, a roadmap, and a risk posture — or whether they are running in isolation, each one re-litigating questions that other projects have already encountered. The framework does not require killing existing projects to bring them into the canvas. It requires the discipline of bringing them onto the canvas as they exist now, and revising the canvas as they run.

Do you have policy-layer fragments built before evidence? Many organizations have AI principles statements, governance committees, ethics frameworks, or compliance regimes that were stood up before any projects produced relevant evidence. These are not necessarily wrong — sometimes they are right by accident — but they are unlikely to fit the actual patterns the organization is producing. The question is whether the existing policy fragments are operating as the framework would want them to, or whether they are constraining work without addressing real friction. Either is possible. The diagnostic is honest review, not automatic dismantling.

Do you have an L&D function already tasked with AI fluency? Most organizations do, in some form — even if the form is one person with a partial allocation. The framework’s posture is to partner with what exists, not to set up parallel learning structures. The diagnostic question is whether the existing L&D function is operating in a push mode (curriculum-out, attendance-tracked) or a pull mode (community-up, demand-driven). The answer is usually push, and the question becomes how the partnership with the AI Enablement Lead can shift the operating mode without dismantling what is already there.

These four questions tell you what you have to work with. The rest of this chapter is about what you put in place from there.

Minimum conditions for the framework to operate

Trellis can operate at almost any organizational scale, but a small set of conditions has to be true for it to operate at all. None of these are sequences. They are conditions — things that need to exist for the framework to produce the kind of evidence accumulation it is built for.

An AI Enablement Lead, or someone playing that role. Title less important than function. This is the person who chairs The Forge, holds the candidate-pattern list, coordinates across the three layers, and partners with the L&D function on capability work. In a small organization this might be a half-time role within an existing function. In a large organization it might be a full-time leadership role with a small supporting team. The framework requires a person; it does not require an organizational chart.

An executive sponsor. The institutional cover the Lead does not have on their own. The sponsor’s job is not to direct the work but to make space for it — to provide air cover when the policy layer is empty and someone outside the framework asks why nothing is being codified, to remove blockers when a project hits a governance pathway at the wrong altitude, to escalate evidence into the strategic loop when a project produces something that warrants strategic revision. A framework without a sponsor will run, but it will run brittle.

A working strategic anchor. Not a polished strategy document. Just enough articulation of what the business is and where it is going that AI candidates can be evaluated against it. If the answers to the four anchor questions are unsettled, the strategic work in Part II becomes the first move. If they are settled enough to make calls, the framework can begin running projects against the existing answers and revise them as evidence comes in.

A venue for project visibility — some version of The Forge. The Forge does not have to be elaborate. In a small organization it is a recurring meeting with a few people. The function is what matters: a place where candidates are gated, canvases are reviewed, and patterns surface. Without this, projects run in isolation and the policy layer never receives input.

An honest partnership with the existing L&D function. The capability work in Chapter 13 cannot be done without the L&D team, and the L&D team cannot do it well without the AI Enablement Lead’s connection to the operational project portfolio. Embed, do not replace.

These are the minimum. Anything else — a balanced portfolio, a populated policy layer, a thoroughly revised strategic loop — is downstream of these conditions being in place.

Patterns to watch for

The framework is built to navigate around a set of recurring failure modes that organizations adopting AI tend to repeat. Each one has a characteristic signature. Each one is worth recognizing as it appears.

Over-planning before evidence. The pull to produce a strategic document before any projects have run is the failure mode this framework was most directly built to prevent. It manifests as a leadership team that spends months perfecting a deck that nobody can argue with and that nobody can actually act on. The discipline is to start running projects as soon as the strategic anchor is workable, not when the document is finished.

Treating governance as a gate rather than an overlay. Governance in a healthy framework is not a checkpoint that projects pass through one time at the start. It is a layer that runs alongside the project — risk questions on the canvas at day one, autonomy thresholds that move with evidence, ethical stances that sharpen as outputs appear. When governance becomes a gate, projects either grind to a halt at the gate or run around it. Neither produces good outcomes.

Automating moments that need to feel human. The temptation to automate every customer interaction or every employee touchpoint is strong, and the technology is genuinely capable. The customer delight question from Chapter 5 — would a customer actually be glad we did this? — is the discipline that catches this failure mode before it ships. Not every workflow improvement is welcome. Brand promise is not always operational efficiency.

Scaling pilots that have not yet produced evidence. A pilot that posts good early numbers in a small population is not yet evidence the same outcomes will hold at scale. The graduation criteria from Chapter 11 are the test. Skipping them — scaling because the pilot looked promising rather than because the criteria were met — is one of the most expensive failure modes in AI adoption. It produces deployments that look successful at first and then reveal the issues nobody had time to surface in the pilot.

Waiting for full readiness before starting anything. The mirror image of premature scaling. Some organizations stall indefinitely waiting for the data infrastructure to be perfect, for the governance framework to be complete, for the cultural envelope to be defined. The framework’s hypotheses-then-hardening principle exists to prevent this. Start with light sketches, run the project, harden the answers as evidence comes in.

These five failure modes do not depend on organization size or starting state. They show up everywhere. Naming them in advance does not prevent them from appearing — it just makes them legible when they do.

Signals it is working

The framework does not produce visible accomplishment in the way leadership teams sometimes expect. What it produces instead is a set of working patterns that, once present, indicate the framework is operating as it should.

Projects are running with current canvases. The single most important signal. A canvas that has been updated within the last review cycle is a project that is actually learning. A canvas that has stopped changing is a project that has stopped running.

The Forge is making decisions, not just talking. Candidates either cross the gate or do not. Canvases either advance, hold, or wind down. The body produces explicit outputs at every meeting. If The Forge is running but not deciding, it has become a status meeting, and the body is doing the wrong work.

Patterns are being tracked, not just noticed. The candidate-pattern list exists, has names and brief descriptions, and grows or shrinks as recurrence holds or fails to. An organization that notices patterns but does not write them down is an organization that will not promote them.

Strategy is updating on evidence, not just on the calendar. When something a project surfaces changes a leader’s understanding of what is possible, the strategic loop reflects that change. When nothing changes, the strategy stays as it was. Either is fine. What is not fine is a strategic loop that updates on schedule regardless of whether evidence has arrived.

The policy layer is filling in slowly. A few promotions over time, each one with a real owner and a clear target. A growing list of standing functions that operate in the background of the project work. A near-empty policy layer is fine when evidence has not yet accumulated. A populated policy layer where nothing is being used is a sign that codification ran ahead of evidence.

Capability work is being pulled, not pushed. The L&D function is running offerings that meet identifiable demand from the workforce, not curriculum that nobody asked for. Forums and meetups have organic energy. Workshops are oversubscribed. Programs have waiting lists.

These signals are not a checklist; they are texture. An organization that has six of the six is operating well. An organization that has three of the six and is still trying is operating as the framework intends. An organization that has none after a period of effort has either misapplied the framework or is solving a different problem.

Closing

Trellis is not a framework for slowness. It is a framework for moving without overcommitting.

Two failure modes sit on either side of healthy AI adoption. At one extreme, overcommitment — choosing a vision before evidence, scaling a pilot before it has produced signal, codifying a rule before the organization has discovered what the rule should be. At the other extreme, paralysis — waiting for the data to be perfect, the governance to be complete, the strategy to be final, before any work begins. Both produce the same outcome, which is no actual learning. Trellis is the path between.

The discipline the framework asks for is calibration. Move fast where speed produces evidence — light canvases, short pilots, quick conversations at The Forge. Move slowly where speed would force premature commitment — strategic revision, policy codification, autonomy increases. The artifacts the framework produces — canvas, surface, candidate-pattern list — are deliberately light, because lightness is what makes speed possible. A canvas sketched in three sentences can be revised in ten minutes. A canvas that has become a thirty-page project plan can no longer be revised at all.

The reward is an organization that learns. A strategy that reflects what is actually true, not what was once aspirational. A project portfolio working on what matters, not on what was politically easy to start. A policy layer that absorbs recurring concerns, not one that anticipates problems the organization has not encountered. A capability function that meets the workforce where it is, not one pushing curriculum nobody asked for.

None of these are achievable through advance planning. None are achievable through endless waiting. They are achievable through the discipline of moving with evidence — committing when the signal is real, holding off when it is not, and trusting that the right answers will emerge from the work.

This is what the framework is for. The rest is just doing the work.