When AI Solves the Big Stuff

Are You Ready for the Next Leap?

AI is exhibiting emergent intelligence. Existing models solve problems in new areas simply by being exposed to more data. One reason is that the physical world has a structure that AI can read like a language.

Google’s Cell2Sentence-Scale model uses this principle. It is the first to create and confirm a new cancer treatment. This is possible because each cell is a representation of its gene names, in descending order of frequency. Cell structure is akin to a language that AI speaks.

Meanwhile, Gemini can compose music rather than copy human artists. ABC and MIDI notation are machine readable. Music becomes another modality, along with text, audio and video. By using existing model sets-up and pouring in raw data, AI is developing capabilities that were neither predicted nor planned.

While developments in maths, science and music may seem far removed from business, recent progress has significant implications. Three areas in particular stand out. These are precise prediction, data as an asset or liability, and machine discovery.

Precise Prediction

Google DeepMind’s GraphCast and newer systems consistently beat traditional numerical weather prediction on most metrics. The Financial Times reported AI forecasting beating conventional methods worldwide. DeepMind’s own paper details accurate forecasts generated in seconds, rather than the hours it takes existing supercomputer models.

Imagine you run a roofing and solar installation firm. Surprise adverse weather disrupts your schedule and margins. With 10-day forecasts that are measurably better, it is possible to book panel deliveries for fair-weather windows, reschedule crews to interior work during storms, and arrange cranes only when probabilities justify it. Rather than trusting your gut, you price, plan and insure jobs with a forecast that’s already been benchmarked against the best traditional systems.

Wherever weather, demand, traffic, or supply timing matters, prediction becomes a utility. If a competitor runs that playbook and you don’t, your costs and on-time rates start to diverge and your margins suffer.

Data as an Asset or Liability

When AI interprets raw data as a language or modality, it can use it to run multiple simulations. Models will infer a lot from public or commoditised data. Any business that relies on data that is not unique, well-governed and first-party, may find its advantage spirited away.

Consider a regional digital agency. Historically, its edge is ad metrics and playbooks. As frontier models absorb open benchmarks and best practices, this generic performance data stops being special.

Meanwhile, the real asset is the structured, first-party data, such as creative variants, audience annotations and sales-qualified signals. This data allows fine-tuning or conditioning of models for client-specific benefits.

If you don’t manage permissions and retention, proprietary data leaks away. The practical move is to build a data vault, keep it clean and make every AI workflow traceable.

Your moat isn’t having data, but having the right data, structured and governed in a way which models cannot easily re-create.

Machine Discovery

DeepMind’s AlphaFold 3 extends beyond proteins to multi-molecule interactions. These include protein-to-protein, protein-to-ligand and protein-to-nucleic acid. While the knowledge is not open-source, the capability is real. Predicting how well molecules fit together used to require months of wet-lab work.

Specialty food manufacturers are seeking cleaner labels. This means simpler, more natural ingredients. In the past, changes to shelf-life meant slow and expensive trials. Now, with partner labs using AF3-class tools, enzyme and binding-agents may be shortlisted based on model predictions of stable texture, before starting an R&D cycle.

In the meantime, the business concentrates its efforts on validating options with a pilot plant, negotiating ingredient sourcing, adjusting processes and communicating changes to retailers. While the model accelerates discovery, humans add value by getting it safely into production, with documentation and claims that pass regulatory and buyer scrutiny.

As discovery becomes automated, small firms win by owning the last mile. This means being expert at integration, compliance and customer trust.

What to Do This Quarter

Adopt prediction where it moves money. If weather, logistics, or the timing of demand cause volatility in your P&L, then pilot AI forecasting and use it to reprogramme scheduling and pricing. Track the results against your old system.

Turn data into a governed product. Build a schema, with permission and retention policy, label provenance and audit trails by default. Then use that data to condition models for your clients or operations.

Position your team as translators. Collaborate with organisations that have frontier discoveries in areas such as materials, bioprocesses and data optimisation. Invest in applying that knowledge through validation, SOPs, compliance and stakeholder communications.

Most people’s experience of AI is as consumer or work tools. These have flaws, but are useful in saving time and removing tedious tasks. Meanwhile AI research is compressing uncertainty and accelerating discoveries. Small businesses that stay abreast of these developments and wire the gains into their operations will set the pace.

Questions to Ask and Answer

1. Am I making the most of the client-specific data in my business?

2. How will I reposition humans when AI does design and discovery?

3. How much prediction goes into my pricing, planning and production?