A design leader who predicts the future confidently is a design leader who is about to be wrong in public. But a design leader who refuses to plan for the future at all is worse, because the tooling and hiring decisions you make in 2026 commit your team to a trajectory that plays out over the next two years whether you planned for it or not. The skill this lesson teaches is the one in between: reading the signals that are actually visible in 2026, naming the genuinely different futures they could resolve into, and building a plan that survives across more than one of them. This is forecasting as a hedging discipline, not prophecy. The artifact is a two-page horizon scan: a short, dated document that names the signals, sketches four plausible 2027-2028 scenarios, and states what your team should do that holds up no matter which one arrives. You will be wrong about which future shows up. The horizon scan makes sure you are not unprepared for the one that does.

Why Forecast at All, and Why It Is Dangerous

The reason a design leader has to do this is that the major decisions of your role have multi-year tails. If you hire three Design Engineers in 2026, you are betting on a particular relationship between design and code holding for the next few years. If you standardize your team on Figma's ecosystem, you are betting on Figma's trajectory. If you build your design system to be read by AI agents, you are betting that agentic workflows mature in the direction the early signals suggest. These are not reversible-in-a-sprint decisions; they are commitments, and a commitment made with no view of the future is a commitment made blind.

The danger is equally real and runs in two directions. Over-commit to a single predicted future and you build a team optimized for a world that does not arrive, like the teams that bet everything on a tool that got acquired and sunset. Refuse to commit to anything and you build a team that is generically prepared for everything and excellent at nothing, perpetually waiting for clarity that never comes. The horizon scan threads this by separating two things most forecasts conflate: the signals, which are real and observable now, and the scenarios, which are uncertain and must be held loosely. You reason rigorously about the first and hedge deliberately about the second.

The Signals Actually Visible in 2026

A horizon scan is only as good as the signals it is built on, and the discipline is to use signals that are observable facts, not vibes from a conference keynote. Here are the ones that are real as of mid-2026 and that a design leader should be reading.

Figma Is Becoming an Agent-Readable Platform

The most concrete signal is the direction of Figma. The Figma MCP server and Code Connect were released on February 17, 2026, which means design files are now something an AI agent can read structurally, not just look at, and design components can be mapped to their real code counterparts. Figma Make, announced at Config 2026, pushes the platform further toward prompt-to-artifact generation inside the design tool itself. Read together, these are not isolated features; they are a coherent bet that the design file becomes a structured source that both humans and agents read and write. The signal is directional and strong: Figma is investing in a future where the boundary between the design file, the code, and the agent that moves between them gets thinner.

Adobe Is Betting on Agentic Creative Workflows

The parallel signal from the other incumbent is Adobe's move toward agentic creative work. The Adobe Firefly AI Assistant and the agentic-workflow direction shown at MAX 2026 point at creative tools where you direct an agent through a multi-step creative task rather than executing each step by hand. Combined with Firefly's commercial indemnification position, the signal is that Adobe is betting the future of creative tooling is agent-directed rather than manual, with the human moving up to direction and the agent handling execution. This is the same directional bet as Figma's, arriving from the visual-creative side rather than the product-design side.

The App Generators Keep Climbing

The third signal is the revenue and adoption trajectory of the prompt-to-app generators. Lovable and Bolt posted the kind of growth in 2025 that does not reverse quietly: Lovable reaching roughly $206M ARR and 8M users, Bolt reaching roughly $40M ARR in months. The signal here is not that any specific tool wins; it is that the market is paying, at scale, for the ability to go from prompt to working application without traditional handoff, and that demand pressure on the design-to-build boundary is sustained, not a fad. A design leader should read this as evidence that the pressure on the traditional design-then-engineer sequence is structural and will keep increasing.

The Design System Itself Is Becoming an Interface for Agents

The fourth signal is more conceptual but no less real: the emerging idea, articulated by practitioners including Brad Frost in his work on agentic design systems, that the design system stops being only a library humans pull from and becomes an interface that AI agents read from to generate conforming work. If agents generate UI, the design system is the thing that constrains them to your standards, which makes a well-structured, machine-readable design system the control surface for AI-generated design. The signal is that the design system is being reframed from a human convenience into governance infrastructure for an agentic future.

Reason rigorously about the signals, which are real and observable now. Hedge deliberately about the scenarios, which are uncertain and must be held loosely. The failure is doing the opposite: treating a confident scenario as fact, or dismissing a strong signal because the future built on it is uncertain.

The Four Plausible Futures These Signals Could Resolve Into

The signals all point in a similar direction, toward agentic, generative, boundary-dissolving tooling, but they are consistent with several quite different worlds. Naming four of them, and being honest that you do not know which arrives, is the core of the scan. The point is not to pick a winner; it is to ensure your plan is robust across the spread.

Future One: The Integrated Platform Wins

In this future, one or two incumbents (most plausibly Figma and Adobe) successfully integrate design, generation, code, and agents into a coherent platform, and the stack consolidates. The design file, the agent, and the code live in one ecosystem, and the winning move is deep investment in that ecosystem: Figma-native workflows, Code Connect everywhere, design systems wired into the platform's agent layer. The risk if you bet on this and it does not arrive is lock-in to an ecosystem that turned out not to be the center of gravity. The tell that this future is arriving is the integrations getting genuinely seamless rather than demo-deep.

Future Two: The Best-of-Breed Stack Persists

In this future, no single platform wins, and teams continue to assemble a stack from specialized tools connected by open protocols like MCP. The design file is in one tool, generation in another, code in a third, stitched together by standards rather than by a single vendor. The winning move here is investing in the connective tissue, MCP literacy, open standards like the design-tokens spec, and an explicitly portable design system, rather than in any one vendor. The risk of betting on this and being wrong is dispersing effort across a stack that a winning platform later makes obsolete. The tell is whether open protocols keep gaining traction or get absorbed into proprietary platforms.

Future Three: The Design-Build Boundary Genuinely Collapses

This is the future the app-generator trajectory most directly suggests: the handoff from design to engineering largely dissolves for a meaningful class of work, because prompt-to-app generation gets good enough that designers (or PMs, or anyone) produce shippable artifacts directly. In this world the Design Engineer role becomes central, the distinction between a prototype and a build erodes, and the highest-leverage skill is directing generation toward production-quality output. The risk of over-betting on this is hollowing out the craft and verification capacity your team needs when the generated output is wrong, which it still frequently is. The tell is whether generated output crosses the line from "impressive demo" to "ships to production without a rebuild" for non-trivial work.

Future Four: The Correction and Consolidation

The fourth future is the one the hype cycle makes hardest to take seriously and that a disciplined forecaster must include: the agentic-tooling wave hits real limits, some heavily-funded tools fail or get acquired and sunset, and the market corrects toward a more sober equilibrium where AI augments but does not transform the workflow as dramatically as 2026 expected. In this future, the teams that kept their craft fundamentals strong and did not over-rotate on any single tool win, and the over-committed teams are stranded. The risk of betting on this is being too conservative and falling behind teams that adopted faster. The tell is funding drying up, consolidation accelerating, and the gap between tool marketing and tool reality becoming a public story.

Building a Plan That Is Robust Across All Four

The payoff of naming four futures is that it forces you to find the moves that are good in all of them, and those moves are the spine of the horizon scan's recommendation. The discipline is to ask, for each candidate decision, in how many of the four futures does this hold up. A move that wins in all four is a no-regret move you should make now. A move that wins in only one is a bet you make only if you have a strong reason to believe that future, and even then you size it so being wrong is survivable.

Several no-regret moves fall out of this analysis, and they are worth naming because they are where you point your 2026 investment. A machine-readable, well-governed design system is valuable in every future: it is the control surface for agents if the platform or boundary-collapse futures arrive, it is the portable asset if the best-of-breed future arrives, and it is solid craft infrastructure if the correction arrives. MCP and open-standard literacy is similarly robust, because it positions you for the best-of-breed world and costs little if a platform wins. And keeping your team's craft and verification fundamentals strong is the ultimate hedge, valuable in every future and decisive in the correction. These are the moves you commit to with confidence because the future does not have to cooperate for them to pay off.

The bets, by contrast, are the moves that depend on a specific future. Going all-in on a single vendor's ecosystem is a bet on Future One. Restructuring the team around the dissolved design-build boundary is a bet on Future Three. You can make these bets, but the horizon scan's job is to label them honestly as bets, size them so a wrong call is recoverable, and define the tells that would tell you the bet is paying off or failing. A bet you have labeled and sized is responsible leadership; a bet you have mistaken for a certainty is the decision that strands your team.

The Discipline of the Tell

The most useful and least practiced part of forecasting is naming, in advance, the observable signal that would tell you which future is arriving. Without pre-committed tells, you will interpret every new data point as confirmation of whatever you already believe. With them, you have a tripwire: if generated output starts shipping to production without a rebuild for non-trivial work, Future Three is arriving and you act on the bet you sized for it. If two heavily-funded generators fail in a quarter, Future Four is arriving and you lean on your craft hedge. The tells convert the horizon scan from a one-time document into a living instrument: you revisit it when a tell fires, not on a calendar. This is what makes the scan a tool for staying oriented rather than a prediction you are stuck defending.

How to Write the Two-Page Horizon Scan

The artifact is deliberately two pages, because a horizon scan that runs to ten pages does not get read and does not get revisited. The structure that works is tight. Page one is the signals and the scenarios: a short, dated list of the observable signals with their sources, followed by the four scenarios stated in a sentence or two each, with the explicit caveat that you do not know which arrives. Page two is the plan: the no-regret moves you commit to now, the labeled bets with their sizing, and the tells you are watching for each scenario. Date it prominently, because the entire document is a snapshot of a moving target and an undated horizon scan is worse than useless, as it gives the false impression of current truth.

Write it for an executive audience, because part of its job is to let you have the tooling-and-hiring conversation with a CPO or CFO from a position of having thought it through rather than reacting to the latest LinkedIn panic. The horizon scan is the document that lets you say, when the CEO forwards you a breathless article about the tool that will replace your team, "we have a view on this; here are the signals, here is the range of outcomes we plan for, and here is why our investments hold up across them." That posture, prepared rather than reactive, is most of what executive credibility on AI strategy actually is.

The Honesty That Makes It Credible

The credibility of the scan rests entirely on its intellectual honesty, and the specific honesty that matters is admitting uncertainty without using uncertainty as an excuse for inaction. A scan that confidently predicts one future is not credible to anyone who has watched the last three years of tool churn. A scan that throws up its hands and says the future is unknowable is not useful. The credible scan does the harder thing: it commits firmly to the signals and the no-regret moves, holds the scenarios genuinely loosely, and is explicit about which of its recommendations are robust and which are bets. That combination, confident where the evidence supports confidence and humble where it does not, is exactly the disposition that separates a design leader people trust on AI strategy from one who is either a hype-merchant or a denier.

Key Takeaways

• Forecasting for a design leader is a hedging discipline, not prophecy. Your 2026 tooling and hiring decisions have multi-year tails, so you must plan for the future even though you will be wrong about which one arrives. Over-committing to one prediction strands your team; refusing to commit builds a team excellent at nothing. The horizon scan threads this by separating signals (reason rigorously) from scenarios (hold loosely).
• The real, observable 2026 signals: Figma is becoming agent-readable (MCP server and Code Connect on February 17, 2026; Figma Make at Config 2026); Adobe is betting on agentic creative workflows (Firefly AI Assistant and the MAX 2026 direction); the app generators keep climbing (Lovable around $206M ARR and 8M users, Bolt around $40M ARR in months); and the design system is being reframed as agent-readable governance infrastructure (Brad Frost's agentic design systems).
• Four plausible futures the signals could resolve into: the integrated platform wins (consolidation around Figma/Adobe), the best-of-breed stack persists (specialized tools connected by open protocols), the design-build boundary genuinely collapses (prompt-to-app good enough to ship), and the correction (the wave hits limits, tools fail, the market sobers). You do not know which arrives; the scan's job is robustness across the spread.
• The no-regret moves win in all four futures and are where you commit 2026 investment with confidence: a machine-readable, well-governed design system (control surface, portable asset, or craft infrastructure depending on the future), MCP and open-standard literacy, and strong craft and verification fundamentals (the ultimate hedge, decisive in the correction).
• The bets depend on a specific future (all-in on one vendor bets on integration; restructuring around the dissolved boundary bets on collapse). Make them only when labeled honestly as bets, sized so a wrong call is recoverable, with pre-committed tells that signal whether the bet is paying off.
• Write the scan as two dated pages: page one signals and scenarios with sources and an explicit uncertainty caveat, page two the no-regret moves, labeled bets, and per-scenario tells. Its credibility rests on admitting uncertainty without using it as an excuse for inaction; confident where evidence supports it, humble where it does not, which is the posture that earns executive trust on AI strategy.