The stories of famous inventions get told as flashes of genius, but the work underneath is almost always the same five-stage loop: define the problem, explore possible solutions, build a prototype, iterate by testing and fixing, and refine at scale. This page walks through five inventions that reshaped the modern world — Edison’s incandescent lamp, the Wright Flyer, the Bessemer converter, Ford’s moving assembly line, and CNC machining — and pulls out the specific stage of the design process each one illuminates best, with a few notes from nearly two decades of watching American small and mid-sized businesses run that same iteration loop on a shop floor.
What the design process is
A design process is a repeatable sequence of steps for turning a problem into a working product. Different organizations draw the boxes slightly differently — Science Buddies, for example, teaches an eight-step student-facing version — but the structure is stable. You define the problem specifically enough that you could tell if you had solved it. You explore the solution space broadly enough that you are not just refining the first idea that came to mind. You prototype, meaning you build something cheap and partial that forces you to confront physical reality rather than argue on a whiteboard. You iterate, which is the long middle in which most of the actual design work happens: you test, the test fails in an interesting way, you adjust, you test again. And eventually you refine at scale, which is a separate problem from inventing the thing at all — a lab prototype does not care about cost per unit, a production design does.
The most common misunderstanding of the design process is that the five stages run strictly in order, like a checklist. They do not. A working prototype that fails in the wrong way sends you back to problem definition. A clever solution you tried to prototype sends you back to exploring alternatives. Iteration is not a stage that happens at the end — it is the connective tissue between every other stage. The inventions below are famous in part because their teams stayed in the iteration loop much longer than was considered reasonable at the time.
The five stages, in one diagram
The five stages each invention on this page moves through — with the return arrow that real design work never stops drawing.
The solid arrows are the textbook order. The dashed return arrow is what every working project actually looks like.
Each of the five inventions below illustrates one stage better than the others. You can read them in order or jump straight to whichever one sounds most useful.
Edison’s incandescent lamp
The carbonized-filament incandescent lamp
Design-process lesson: Define the problem specifically enough to test thousands of candidate answers against it.
The usual textbook framing — Edison “invented the lightbulb” in a flash of insight — is almost exactly backwards. Incandescent lamps existed before Menlo Park; several inventors on both sides of the Atlantic had produced devices that glowed for short periods. What Edison’s team defined specifically, and what earlier attempts had not, was the commercial-viability problem: find a filament that could be heated to incandescence in a vacuum for long enough, and cheaply enough, to displace gas lighting in ordinary homes.
On October 21–22, 1879, after months of work on platinum filaments that kept failing because the hot metal weakened too quickly in air, Edison’s staff at Menlo Park ran a now-famous experiment with a filament of carbonized cotton thread inside a sealed vacuum bulb. It burned long enough to convince them the approach was right. According to the Thomas A. Edison Papers archive at Rutgers, the laboratory then tested an extraordinary breadth of carbonizable materials — woods, papers, vulcanized fibre, celluloid, lampwick, flax, cork, coconut hair and shell, fishing line, and eventually grasses and canes. After the broad material search, carbonized bamboo became the standard Edison filament.
Edison is often quoted — accurately or not — as saying he found many thousands of ways that did not work. Whatever he said, the archival record is consistent with the claim. The reason Edison’s version of the incandescent lamp won is not that he had better intuition than everybody else. It is that he had a tighter problem definition and an iteration budget the size of a small manufacturing company, and he kept running the loop.
The Wright Flyer
Powered, sustained, controlled flight
Design-process lesson: Build your own measurement tools when the published data is wrong.
Wilbur and Orville Wright were bicycle-shop owners in Dayton, Ohio, with no formal engineering training. They started working seriously on flight in 1899 and made their first powered flight at Kitty Hawk, North Carolina, on December 17, 1903. The four years in between are the interesting part of the story for anyone trying to learn how design actually works.
By the end of 1901, the Wrights were frustrated. Their 1900 and 1901 gliders were generating far less lift than the standard published aerodynamic tables — especially those compiled by the German glider pioneer Otto Lilienthal, whom the Wrights had treated as the reference of the day — said they should. Most teams in that situation would have kept redesigning the glider. The Wrights did something harder: they questioned the data. In the back of their Dayton, Ohio bicycle shop, they built a small, single-speed, open-return wind tunnel — a long wooden box with a fan driven off the shop’s gas engine by a belt drive. Through the fall of 1901, as NASA Glenn’s educational archive of the tunnel work documents, the brothers cut airfoil shapes out of 20-gauge sheet steel and measured lift and drag themselves on between one and two hundred models. The tunnel data revealed that several of Lilienthal’s coefficients were simply wrong for the wing profiles the Wrights were using.
The 1902 glider, designed around the Wrights’ own numbers, flew under full three-axis control. The 1903 Flyer that followed was an incremental refinement with an engine on it. The general lesson travels well: when a prototype disagrees with published data, do not assume the prototype is wrong.
The Bessemer converter
Mass-produced steel from molten pig iron
Design-process lesson: When a working prototype fails in production, the answer is sometimes not the machine.
Before the Bessemer process, steel was made in small batches by crucible methods — slow, expensive, and unable to meet the demand of railroad builders and structural-iron buyers. Henry Bessemer’s 1856 breakthrough, patented in England, was a converter: a pear-shaped cylindrical vessel, about 20 feet tall, into which molten pig iron was poured and through which air was blown. The oxygen in the air burned off silicon, manganese, and carbon impurities as slag and, in doing so, raised the temperature of the remaining iron high enough to keep it molten as it became steel. Bessemer demonstrated the process publicly, his bench results were spectacular, and licensees lined up.
Then the first licensee batches failed. The converter produced a brittle, unworkable metal. Bessemer’s reputation took a serious hit, and for several years the process was considered discredited.
The mechanical design was fine. The failure was chemical. Bessemer’s own test pigs had happened to be low in phosphorus; most licensee iron was not, and phosphorus contamination destroyed the finished steel. The practical fix came from several contributors working on the metallurgy rather than on the converter. Robert Forester Mushet discovered that adding an alloy of carbon, manganese, and iron — spiegeleisen — after the air blow restored the correct carbon content and scavenged sulfur. The Swedish ironmaster Göran Göransson redesigned the converter’s operation so it ran reliably in practice. With those corrections, the Bessemer process became the first method for mass-producing cheap steel, and steel replaced iron in rails, bridges, and eventually whole buildings. The design lesson is uncomfortable: a prototype that works is not the same as a design that works everywhere, and the fix is sometimes not in the part you are staring at.
Ford’s moving assembly line
Continuous flow production of the Model T
Design-process lesson: Study a process that already works before designing anything new.
By 1913, the pieces of what would become the moving assembly line already existed. Ransom Olds had used a stationary assembly line for the curved-dash Oldsmobile since 1901. Interchangeable parts had been refined in American armories for most of the 19th century. Division of labor was centuries old. What did not yet exist was a continuously moving assembly of an entire automobile on one synchronized line. See our History of the Assembly Line for the full arc.
The design decision worth pulling out here is the one Ford’s engineers made before they drew any fixtures: they went to Chicago and Cincinnati and watched the meatpacking industry’s disassembly lines. Carcasses rode an overhead trolley past fixed stations where specialized butchers made one cut each. Flow, division of labor, and specialized workstations were already running at industrial scale in the food industry. Ford’s team extracted the pattern, reversed the direction, and applied it to assembling a car. Ford himself described what they had built as “the disassembly line run in reverse.”
The first experiment, in April 1913, was narrow: flywheel-magneto sub-assembly on an overhead chain. Productivity roughly doubled. Over the next eighteen months, the moving-line concept was extended to engines, transmissions, and finally the full chassis at the Highland Park plant. According to The Henry Ford, total time to assemble a Model T fell from roughly 12½ hours per car under the older stationary method to about 93 minutes per car once the moving line was fully implemented — an improvement of roughly eightfold. The lesson here is a variant of the other inventors’ lesson: most of the original thinking happened in problem definition and scouting. By the time Ford’s engineers started designing new equipment, they already had a working reference model — in a different industry, for a different purpose — to copy from.
CNC machining
Numerically controlled milling machines
Design-process lesson: A well-defined industrial problem can drive the development of the tool that solves it.
In the late 1940s, the U.S. Air Force was trying to produce aircraft parts that manual machinists could not reliably cut: continuously curved helicopter rotor blades, integrally stiffened aircraft skins, and the compound-curve control surfaces of jet-age airframes. The parts needed to be interchangeable across units and accurate to thousandths of an inch, and the geometry was beyond the reach of a human operator turning a hand wheel while reading a paper drawing.
John T. Parsons, a Michigan subcontractor who had been trying to produce helicopter blade templates for the Air Force, proposed a specific answer: drive a milling machine’s cutting head from a table of numerical coordinates punched onto cards, rather than from a human hand on a crank. Parsons worked with engineer Frank Stulen on the concept. Through the late 1940s and into the 1950s, the U.S. Air Force contracted the Servomechanisms Laboratory at MIT to develop an experimental numerically controlled milling machine, and MIT demonstrated a working machine on that program. The National Inventors Hall of Fame credits Parsons with the invention of numerical control and names Stulen as his collaborator. The U.S. Patent and Trademark Office’s National Medal of Technology and Innovation record confirms that in 1985 Parsons and Stulen received the National Medal of Technology for their development and demonstration of the numerically controlled machine tool.
Modern CNC machining — with digital G-code in place of the punched cards, and today a five-axis machining center that can hold a part in one setup while the spindle reaches around to cut it from five directions — is the direct descendant of that design chain. What makes it a clean case study of the design process is the sequence: first, a specific industrial problem defined tightly enough that it pointed at a particular kind of tool; then, a prototype built by a research lab; then, decades of iteration across academia, the aerospace industry, and tool manufacturers; and finally, refinement at a scale that put a numerically controlled machine into every job shop in the country. The math problem found the machine, not the other way around.
A lender’s view of iteration
I spend most of my working hours at Crest Capital structuring financing for the machines that American small and mid-sized businesses use to iterate on their own products — CNC mills, printing presses, welding cells, packaging equipment, diagnostic imaging. From that seat, a few patterns in the design process are worth pulling out.
Edison’s Menlo Park laboratory was funded by a mix of Edison’s own patent royalties and backers who bought stock in the company he formed around the lighting work. The Wrights self-funded the 1899–1903 iteration out of their bicycle shop’s cash flow. Bessemer licensed his process for ongoing royalties once the metallurgy was fixed. Ford funded Highland Park largely from the Model T’s operating cash. The MIT Servomechanisms Lab’s numerical-control work was a U.S. Air Force research contract. The through-line is not the specific financing structure — those change every generation. The through-line is that somebody has to pay for the iteration loop before the design stops being an idea and starts being a thing. Today a shop iterating on a production process typically funds the equipment side with a three-to-seven-year equipment lease or term loan. The deal structure is different from Edison’s; the role it plays in the design process is the same.
Here is one I see play out a lot in American machine shops. A family-run CNC shop finances a new five-axis machining center to replace an older three-axis machine. Month one, the shop runs the new machine with the same workholding they used on the three-axis — the chuck, a manual vise, a set of parallels. It works, but the shop is using maybe half of what the machine is capable of. Month two, the lead machinist sketches out a tilting fixture that lets the spindle reach angles the vise was blocking; they cut it in-house on the new machine itself. Month four, they add a tool-setting probe, which cuts setup time materially on short-run jobs. Month eight, they add a pallet changer so the operator can load the next job while the current one is still cutting. By the time a year has gone by, the shop is running work nobody on the floor could have specified on day one. From a lender’s seat, the interesting thing is that the capital decision de-risked itself over time through the iteration, not through the original spec. The design process did not stop when the machine arrived. That is the year when a shop actually learns what it bought.
When I underwrite a deal, one of the things I pay attention to is how a shop talks about its own process. A shop that can describe the specific step in its workflow it is trying to speed up, and name the equipment it believes will speed it up, is a shop that has been running its own version of the design loop on its production line. Most of the time, by the time such a shop is calling Crest about a new machine, it has already iterated on its existing process for months — tweaked tooling, resequenced work cells, retrained operators — and it is bringing me the one problem it cannot solve without capital. Underwriters love those deals, because the thinking behind them is already visible. Edison, the Wrights, Bessemer’s metallurgists, Ford’s engineers, and John Parsons would all recognize the posture. Define the problem, exhaust the cheap iterations first, then spend the capital.
Frequently asked questions
What is the design process?
The design process is the sequence of steps an inventor or engineer uses to turn a problem into a working product. Most modern formulations share the same five stages: define the problem, explore possible solutions, prototype, iterate by testing and redesigning, and refine at scale. The process is not linear. Every real design loops back to earlier stages, often many times, before the final product is considered done. Science Buddies’ Engineering Design Process is a widely used student-facing version of the same idea.
How long did it take Edison to invent the incandescent lightbulb?
Edison’s team at Menlo Park worked on the incandescent lamp for roughly a year and a half before their October 1879 carbonized-cotton-thread experiment gave them a lamp that burned long enough to be commercially viable. Over the following months, the laboratory tested many other carbonizable materials — woods, papers, fibers, cork, grasses, and eventually bamboo. Carbonized bamboo, which Edison’s researchers settled on after testing thousands of natural samples, became the standard filament in Edison bulbs for years afterward. Edison himself summarized the pattern as having found thousands of materials that did not work — which is what the design process, told honestly, usually looks like.
Why did the Wright brothers build their own wind tunnel?
By the end of 1901, the Wright brothers’ gliders were generating far less lift than the published aerodynamic tables — especially Otto Lilienthal’s, which were the standard reference of the day — predicted they should. Rather than keep refining their aircraft against data they suspected was wrong, they built a small, single-speed wind tunnel in the back of their Dayton, Ohio, bicycle shop and measured lift and drag themselves on between one and two hundred miniature airfoil shapes they cut from sheet steel. The tunnel data reshaped their designs, and the 1902 glider flew under control. The lesson is that when published data is in conflict with what a prototype is actually doing, the published data is sometimes the thing that needs fixing.
Why did the first public Bessemer converter fail?
Henry Bessemer’s 1856 demonstration of his steel-making converter was successful on his own pig iron, but early licensees produced brittle, unusable metal with the same vessel. The converter and the air-blowing principle were sound; the problem was chemistry. Bessemer’s original test iron happened to be low in phosphorus, but the iron most licensees used was not — and phosphorus contamination ruined the finished steel. The practical fix came from several contributors, including Robert Forester Mushet, whose post-blow addition of a manganese alloy restored the right carbon content and scavenged sulfur, and Swedish ironmaster Göran Göransson, who redesigned the converter’s operation. In other words, the breakthrough was metallurgical refinement on top of a working mechanical prototype, not a new vessel.
How did Henry Ford’s engineers design the moving assembly line?
Ford’s engineers at the Highland Park plant studied flow production in places that already worked before designing any new equipment: Chicago and Cincinnati meatpacking plants, where carcasses rode an overhead trolley past fixed stations of specialized butchers. Their first experiment, in the spring of 1913, was a moving flywheel-magneto assembly on an overhead chain — a narrowly scoped pilot, not the whole car. That pilot roughly doubled productivity, and the concept was then extended over the following eighteen months to the Model T’s engine, transmission, and finally the full chassis. According to The Henry Ford, total assembly time per car fell from roughly 12½ hours per car under the older stationary method to about 93 minutes per car once the moving line was fully implemented.
How did CNC machining get invented?
Numerical control grew out of a specific manufacturing problem in the late 1940s: producing the complex, continuously curved aluminum skins that modern aircraft and helicopter rotors required, at the tolerances the U.S. Air Force needed. John T. Parsons, a Michigan subcontractor, proposed driving a milling machine’s cutting head from a table of numerical coordinates punched onto cards rather than from a human operator’s hands on a crank. Parsons worked with engineer Frank Stulen on the concept, and the U.S. Air Force contracted the MIT Servomechanisms Laboratory to develop and demonstrate an experimental numerically controlled milling machine. The National Inventors Hall of Fame credits Parsons with the invention of numerical control, and the U.S. Patent and Trademark Office’s National Medal of Technology and Innovation record confirms that Parsons and Stulen received the National Medal of Technology in 1985 for their work. Modern CNC machining — with digital G-code where Parsons’s punched cards used to be — is the direct descendant of that design chain.
Selected sources
- Science Buddies — Engineering Design Process Steps A widely used K–12 and undergraduate framework for the engineering design process, covering problem definition, research, requirements, brainstorming, prototyping, testing, redesign, and communication of results.
- Thomas A. Edison Papers (Rutgers) — The Carbon-Filament Lamp Rutgers’s archival essay on Edison’s October 1879 carbonized-cotton-thread experiment and the Menlo Park laboratory’s broad subsequent search through available carbonizable materials.
- NASA Glenn Research Center — Wright 1901 Wind Tunnel NASA Glenn’s K–12 educational account of the Wright brothers’ 1901 wind-tunnel work, including their frustration with the 1900 and 1901 glider tests, their decision to question Lilienthal’s published data, the open-return tunnel built in their Dayton bicycle shop, and their lift-and-drag tests on between one and two hundred sheet-steel airfoil models.
- Encyclopaedia Britannica — Bessemer process Britannica’s entry on the Bessemer process: converter design, the role of air blowing in oxidizing impurities, the early phosphorus failure, and the contributions of Mushet and Göransson that made the process commercially viable.
- The Henry Ford — Henry Ford: Assembly Line (expert set) The Henry Ford’s curated expert set on the moving assembly line at Highland Park — photographs, artifacts, and archival material documenting the 1913 moving line and what it replaced.
- The Henry Ford — K-12 Research Guide: Assembly Line The museum’s research-guide summary of the Highland Park moving line, including the 12½-hour to 93-minute assembly-time figure cited on this page.
- National Inventors Hall of Fame — John T. Parsons NIHF’s biographical summary of Parsons, credited with the invention of numerical control, and his collaboration with aircraft engineer Frank Stulen.
- U.S. Patent and Trademark Office — National Medal of Technology and Innovation, 1985 Laureates The USPTO’s National Medal of Technology and Innovation record for the 1985 laureates, recognizing John T. Parsons and Frank L. Stulen for their development and demonstration of the numerically controlled machine tool for producing three-dimensional shapes.