Optioneering with AI

Optioneering is simply the disciplined generation and evaluation of many design options. The word captures a mindset shift: instead of committing early to one idea and refining it, you hold the space open, explore widely, and let evidence about performance inform the choice. Architects and engineers have always done a version of this by hand - sketching alternatives, comparing schemes - but human time caps it at a handful of options, and the evaluation is often qualitative.

Three ingredients let AI blow that cap off. First, a parametric model - in Grasshopper, Dynamo or a tool like Forma - that can generate variants automatically as you change inputs. Second, fast evaluation of each variant against your goals, which is exactly where the surrogate models of Lesson 7.2 come in: they make scoring thousands of options feasible because each score costs milliseconds, not hours. Third, a search strategy - often an optimisation algorithm like a genetic algorithm - that intelligently roams the design space toward better options rather than trying everything blindly.

Put together, these turn 'explore a few options' into 'explore the space'. This is the natural culmination of the whole module: prediction gave you fast feedback, surrogates made that feedback near-instant, and optioneering harnesses that speed to search systematically. It is generative and computational design in the service of performance - with, crucially, you still holding the objectives and the final call.

The mechanics follow a clear loop. You define the variables (what can change - window ratio, column grid, floor count, layout parameters), the objectives (what to improve - maximise daylight, minimise energy and cost), and the constraints (what must hold - minimum area, code limits, site boundary). The AI-driven search then generates a variant, evaluates it against every objective, and uses the scores to decide what to try next, iterating toward better-performing designs.

Defining these well is the real skill, and it is pure design judgement dressed in technical clothing. Choose the wrong objectives and you will optimise brilliantly for the wrong thing - the classic trap of getting exactly what you asked for and not what you wanted. If you optimise a facade purely for daylight, you may get glare and overheating; if you chase minimum cost alone, you can destroy quality. Good optioneering means encoding a balanced set of what matters, and staying alert that many things you care about - delight, context, how a space feels to move through - resist being turned into a number at all.

A plain-English objective set might read:

VARY:      window-to-wall ratio, orientation, shading depth
MAXIMISE:  useful daylight hours
MINIMISE:  annual energy use, construction cost
KEEP:      net floor area >= brief, glare within limits

That little block is a design brief in disguise. The AI will honour it literally and tirelessly - which is exactly why what you choose to put in it, and what you leave out, is the most important move you make.

It is easy to picture optioneering only as big-form massing optimisation, but the same logic applies from the scale of a city block down to a single room, and recognising that widens where you can use it. At the urban and massing scale, you might sweep building footprints, heights and orientations for daylight to neighbours, self-shading and solar access - the territory of tools like Forma. At the building scale, facade design is a classic case: varying glazing ratio, shading depth and orientation against daylight, glare, energy and cost is a multi-objective problem almost tailor-made for a Pareto front.

At the structural and systems scale, engineers optioneer grids, spans and member sizes against material use, cost and buildability. And at the interior and layout scale, the variables become furniture arrangements, partition positions or workstation counts, scored on circulation, daylight access, capacity and cost - increasingly served by generative-layout tools that produce and rank many plans. The scale changes; the pattern does not: variables, objectives, constraints, search, trade-off, choice.

Two cautions travel across all scales. First, the evaluation must actually mean something at that scale - a daylight metric that makes sense for an office floor may be meaningless for an urban block, and a Pareto front built on a mismatched or out-of-range surrogate is a confident illusion. Second, more options is not automatically better: ten thousand variants you never truly look at can crowd out the reflective judgement that good design needs. The goal is not maximum options for their own sake but better-informed choices - a wide, well-evaluated field that sharpens your decision rather than drowning it. Used at the right scale with honest evaluation, optioneering turns 'I explored what I had time for' into 'I explored what mattered' - which is a real upgrade to how confidently you can defend a scheme.

Here is the idea that makes optioneering genuinely useful rather than naive: real design goals conflict, so there is rarely one best option. More daylight often means more glazing, which can mean more energy and cost. Cheaper usually fights better. When objectives compete, there is no single winner - there is a set of options where you cannot improve one goal without sacrificing another. That set is called the Pareto front, and it is the real output of good optioneering.

Understanding the Pareto front changes how you read results. Every option on the front is 'optimal' in the sense that nothing beats it on all goals at once; options off the front are simply worse - beaten on every count - and can be discarded. But among the options on the front, the machine has nothing to say about which is best, because that depends on how you value daylight against cost against energy for this project, this client, this context. The AI narrows a vast space down to a curated set of genuinely good, genuinely different trade-offs. That is an enormous service - and it is precisely where its authority ends.

This is why the honest framing is 'the machine ranks; the designer chooses'. Optioneering does not remove judgement - it sharpens what judgement is spent on. Instead of grinding through geometry and arithmetic, you spend your expertise on the decision that actually needs it: standing in front of a Pareto front of excellent options and deciding which balance is right. That is a better use of a designer than any the tool could replace.

For all its power, optioneering carries real traps, and naming them is how you stay its master rather than its servant. The first is optimising the wrong thing: the tool perfectly serves whatever objectives you gave it, so a narrow or careless objective set produces a technically optimal, humanly poor result. The guard is to keep revisiting what you are asking for, and to remember the unquantifiable goals the search cannot see. The second is false precision: options are only as good as the evaluation behind them, and if that evaluation is a surrogate, all of Lesson 7.2's caveats apply - a beautifully ranked Pareto front built on out-of-range predictions is a beautifully ranked guess. Validate the finalists.

The third trap is abdication - letting the existence of an 'optimal' option quietly make the decision for you. It is seductive to point at the algorithm's top pick and call it settled, but that outsources the one thing you must own. Use the Pareto front as input to your decision, not as the decision. Bring back the things it never scored: how the winning option sits in its context, what it will feel like, whether it says what the project should say.

Done with that discipline, optioneering is one of the most empowering AI workflows in design. It lets a small team explore like a large one, grounds intuition in evidence, and frees your best thinking for the choices that matter. The pattern is the same one the whole course has built toward: AI does the fast, wide, tireless work - generating, scoring, ranking - and you do the human work of setting the goals and making the call. Explore ten thousand options if you can; just be the one who chooses among the best.