Several technologies available today are strong candidates for being the forerunners of the next major leap in human capability. The most likely among them are large language models (early steps toward general artificial intelligence), CRISPR gene editing (rewriting biology at will), quantum computing (solving problems classical computers never will), and nuclear fusion (virtually limitless clean energy). Each exists in a working but early form, much like the internet in 1993 or powered flight in 1910. Here’s where each one actually stands and why it qualifies.
Large Language Models as a Path to General AI
Today’s AI systems are narrow. They predict text, generate images, or classify data, but they don’t understand the world the way a person does. Large language models are the closest thing we have to a forerunner of artificial general intelligence, the kind of system that could reason flexibly across any domain. The gap between where these models are now and true general intelligence is still enormous, but the trajectory is what matters.
The biggest bottleneck is efficiency. Inference costs, the expense of actually running a model after it’s been trained, have already surpassed training costs and are now the primary barrier to wider deployment. There’s growing pressure to run these models on personal devices like phones and laptops, which demands far more compact architectures. Techniques like pruning (removing less important parts of a neural network) and quantization (reducing the precision of calculations) can shrink models, but both degrade performance. Compressed models consistently fall short of their full-sized counterparts in capability. The scaling path that got us this far, simply making models bigger, is hitting diminishing returns.
A parallel development could change the equation. Neuromorphic chips, hardware designed to mimic how biological neurons fire, use a fraction of the energy conventional processors require. The human brain handles all its cognitive tasks on roughly 20 watts. IBM’s NorthPole neuromorphic chip classified images using a tiny fraction of the energy a standard system needed, and it was five times faster. In April 2025, researchers demonstrated the first large language model running on Intel’s Loihi 2 neuromorphic chip. It matched the accuracy of a conventional GPU-based model while using half the energy. If neuromorphic hardware matures, it could remove the energy and cost constraints that currently limit AI’s growth.
CRISPR Gene Editing in Human Medicine
CRISPR is the first technology that lets scientists edit DNA with enough precision, speed, and affordability to be practical at scale. It has already crossed from laboratory tool to approved medical treatment. A CRISPR-based therapy for sickle cell disease is now in clinical use, making it the first gene-editing treatment to reach patients for a genetic blood disorder. CRISPR-edited crops are also in production, including a more nutritious tomato and high-yield, disease-resistant wheat varieties.
What makes CRISPR a forerunner rather than a finished product is that the current generation of edits is relatively simple: snipping a gene at one location or swapping a single DNA letter. The next frontier is base editing and prime editing, techniques that can rewrite genetic code with far greater subtlety. As these tools mature, the range of treatable genetic conditions will expand from a handful to potentially thousands. The technology that exists today is the Wright Flyer. The 747 is the ability to correct complex multi-gene diseases, engineer drought-proof crops for a warming climate, and perhaps slow biological aging.
Quantum Computing Reaches a Turning Point
Quantum computers exploit the physics of subatomic particles to perform certain calculations exponentially faster than any classical machine. For most of their history, they’ve been too error-prone to do useful work. That’s changing quickly. Quantinuum and Microsoft recently extracted 48 error-corrected logical qubits from just 98 physical qubits, a ratio that would have seemed impossible a few years ago. The logical circuits operated with error rates 800 times lower than the physical circuits they were built from, and in every test performed, the logical qubits outperformed their physical counterparts by a factor of 10 to 100.
This matters because error correction is the single biggest obstacle between today’s experimental quantum machines and the fault-tolerant quantum computers that could transform drug discovery, materials science, cryptography, and financial modeling. Previous estimates suggested you’d need thousands of physical qubits to produce a single reliable logical qubit. Getting 48 from 98 compresses the timeline dramatically. Quantum computing is a forerunner because the core mechanism now works. The engineering challenge is scaling it up, not proving it’s possible.
Nuclear Fusion Keeps Breaking Records
Fusion power, the process that fuels the sun, promises nearly limitless energy with no carbon emissions and minimal radioactive waste. For decades, fusion experiments consumed more energy than they produced. That barrier fell in December 2022, when the National Ignition Facility produced 3.15 megajoules of fusion energy from 2.05 megajoules of laser input. Since then, the gains have accelerated. By April 2025, NIF achieved a record yield of 8.6 megajoules with a target gain of 4.13, meaning the fusion reaction produced more than four times the energy delivered to the fuel capsule. NIF has now achieved fusion ignition ten separate times.
These are laboratory demonstrations, not power plants. The laser system itself consumes far more electricity than the fusion reaction produces when you account for the full facility. But the core physics question, whether controlled fusion can generate net energy, has been answered. The remaining work is engineering: building reactor designs that can sustain fusion reactions continuously rather than in single pulses, and capturing that energy as usable electricity. Multiple private companies and international projects are now racing toward that goal.
Reusable Rockets and the Cost of Space Access
Getting anything to orbit has historically cost tens of thousands of dollars per kilogram. SpaceX’s Falcon Heavy brought that down to about $1,400 per kilogram. The Starship system, currently in testing, is designed to slash costs by another order of magnitude or more. Even as an expendable vehicle, Starship could deliver payloads to low Earth orbit for $250 to $600 per kilogram. With just five or six reuses of both the booster and upper stage, models project costs dropping below $100 per kilogram. At 50 or more flights per vehicle, the math reaches $15 to $19 per kilogram.
Cheap, routine access to orbit is a forerunner technology for everything from space-based manufacturing and solar power to asteroid mining and Mars colonization. The entire economics of space activity changes when launch costs drop by 100x. It’s comparable to how container shipping transformed global trade: the technology isn’t glamorous, but it enables everything else.
Solid-State Batteries and Energy Storage
Standard lithium-ion batteries top out around 250 to 300 watt-hours per kilogram in commercial cells. Solid-state batteries, which replace the liquid electrolyte with a solid material, have demonstrated energy densities of 280 to 410 watt-hours per kilogram depending on the design, with a theoretical ceiling that’s higher still. The volumetric density advantage is even more striking, reaching 560 to 820 watt-hours per liter.
Higher energy density means longer range for electric vehicles, lighter drones, and more compact grid storage. Solid-state designs also reduce fire risk since they eliminate the flammable liquid electrolyte. The forerunner aspect here is that the chemistry works but manufacturing at scale remains difficult and expensive. Whoever solves the production problem unlocks a step change in how we store and transport energy, which ripples into transportation, aviation, portable electronics, and renewable energy grid stability.
Brain-Computer Interfaces
Devices that translate brain activity into digital commands have moved from science fiction to human trials. Neuralink’s implant uses 64 ultrathin wires carrying 1,024 electrodes, giving it significantly higher bandwidth than earlier brain-computer interfaces. Synchron’s Stentrode, a less invasive device inserted through a blood vessel, has been implanted in five patients with severe upper-limb paralysis, with 12-month follow-up data showing sustained use.
Current applications focus on restoring movement and communication for people with paralysis. The forerunner potential extends much further: direct brain-to-computer communication could eventually change how humans interact with every digital system. The resolution and bandwidth of today’s implants are crude compared to the brain’s full signaling capacity, but they’ve proven the core concept that thoughts can reliably control external devices.
3D Bioprinting of Living Tissue
The ability to print functional human organs would eliminate transplant waiting lists and transform medicine. Researchers at Harvard’s Wyss Institute have 3D-printed blood vessels with multiple cell layers that, after seven days of perfusion, showed a threefold decrease in permeability compared to unlined vessels, indicating functional vessel walls had formed. Printed cardiac tissue began beating synchronously after five days and responded correctly to standard heart drugs, speeding up with isoproterenol and stopping with blebbistatin.
Vascularization, growing a network of blood vessels throughout printed tissue, has been the central obstacle in bioprinting. Without blood supply, printed tissue dies within millimeters of the surface. These results represent the early proof that printed vasculature can sustain living, functional tissue. Full organ printing remains years away, but the enabling technology is no longer theoretical.
Why “Forerunner” Is the Right Word
Each of these technologies shares a common trait: the fundamental proof of concept is settled, but the engineering required to reach full potential is still underway. Fusion produces net energy but not grid power. Quantum computers correct errors but can’t yet outperform classical machines on real-world problems at scale. CRISPR cures one disease while thousands more remain targets. The forerunner stage is the period between “it works” and “it works for everyone.” History suggests this gap takes 10 to 30 years to close, and every technology listed here is somewhere inside that window.