This chapter explores how AI may evolve from a tool that assists with hardware design into a co-inventor of computer architectures, using familiar examples such as the Apple A19 Pro, a 3 nm chip, and 12 GB of RAM in the iPhone 17 Pro Max.
The key issue is not simply that AI can design chips faster, but the limitations surrounding accuracy, safety, and result verification. If a system makes a mistake at the architectural level, the effects could extend to real devices, from performance to power consumption. This article therefore examines the opportunities, risks, and future of the chip industry from all angles.
This chapter explores how AI may evolve from a tool that assists with hardware design into a co-inventor of computer architectures, using familiar examples such as the Apple A19 Pro, a 3 nm chip, and 12 GB of RAM in the iPhone 17 Pro Max.
The key issue is not simply that AI can design chips faster, but the limitations surrounding accuracy, safety, and result verification. If a system makes a mistake at the architectural level, the effects could extend to real devices, from performance to power consumption. This article therefore examines the opportunities, risks, and future of the chip industry from all angles.
When AI Does More Than Use Chips—it Starts Inventing Them
This reflects a major turning point in the industry, as AI is moving from being a coding assistant to becoming a tool that can help devise chip structures itself. The work of Anna Goldie and Azalia Mirhoseini suggests that AI may help uncover options humans overlook, although the results still require serious verification.
Consider the Apple A19 Pro (3 nm) in the iPhone 17 Pro Max: the story behind a smaller chip is not just about speed, but also heat, battery life, and stability. If AI makes a design mistake, the impact can immediately reach real-world use.
When AI Does More Than Use Chips—it Starts Inventing Them
This reflects a major turning point in the industry, as AI is moving from being a coding assistant to becoming a tool that can help devise chip structures itself. The work of Anna Goldie and Azalia Mirhoseini suggests that AI may help uncover options humans overlook, although the results still require serious verification.
Consider the Apple A19 Pro (3 nm) in the iPhone 17 Pro Max: the story behind a smaller chip is not just about speed, but also heat, battery life, and stability. If AI makes a design mistake, the impact can immediately reach real-world use.
When Chip Design Can No Longer Keep Up with AI’s Speed
Engineering teams are facing challenges that are changing increasingly quickly, including computational patterns, model sizes, and the tasks AI must handle. Conventional hardware may therefore fail to keep up with new demands, even if it still performs well for its original purpose.
The important question is: if AI can help discover chip structures and test alternatives on its own, how much faster can the design cycle become? The turning point is not about letting AI make every decision for humans, but about opening up new possibilities and having engineers verify whether they are feasible, stable, and suitable for the intended use.
When Chip Design Can No Longer Keep Up with AI’s Speed
Engineering teams are facing challenges that are changing increasingly quickly, including computational patterns, model sizes, and the tasks AI must handle. Conventional hardware may therefore fail to keep up with new demands, even if it still performs well for its original purpose.
The important question is: if AI can help discover chip structures and test alternatives on its own, how much faster can the design cycle become? The turning point is not about letting AI make every decision for humans, but about opening up new possibilities and having engineers verify whether they are feasible, stable, and suitable for the intended use.
Where Ricursive Intelligence Fits in the AI and Chip Ecosystem
Ricursive Intelligence sits between AI model developers and teams designing specialized computer systems. It uses AI to explore hardware designs suited to real-world tasks rather than replacing chip manufacturers responsible for factory production.
Under the vision of Anna Goldie and Azalia Mirhoseini, the company functions as a bridge between software and hardware. AI models need chips suited to their workloads, while chip manufacturers need designs that can actually be built and support long-term use.
Where Ricursive Intelligence Fits in the AI and Chip Ecosystem
Ricursive Intelligence sits between AI model developers and teams designing specialized computer systems. It uses AI to explore hardware designs suited to real-world tasks rather than replacing chip manufacturers responsible for factory production.
Under the vision of Anna Goldie and Azalia Mirhoseini, the company functions as a bridge between software and hardware. AI models need chips suited to their workloads, while chip manufacturers need designs that can actually be built and support long-term use.
From Humans Defining Everything to AI Joining the Design Process
The traditional approach has humans establish the architecture and then manually fine-tune it point by point. This works well when the scope is clearly defined, but becomes time-consuming as the number of options increases.
The new approach allows AI to explore hardware structures, algorithms, and multiple alternatives simultaneously. Humans still define the goals, assess feasibility, and make the final decisions.
| Factor | Traditional approach | New approach |
|---|---|---|
| Primary designer | Humans | Humans working with AI |
| Exploring alternatives | Adjusting one point at a time | Exploring multiple alternatives simultaneously |
| Human role | Defining and adjusting the design | Defining goals and reviewing the design |
From Humans Defining Everything to AI Joining the Design Process
The traditional approach has humans establish the architecture and then manually fine-tune it point by point. This works well when the scope is clearly defined, but becomes time-consuming as the number of options increases.
The new approach allows AI to explore hardware structures, algorithms, and multiple alternatives simultaneously. Humans still define the goals, assess feasibility, and make the final decisions.
| Factor | Traditional approach | New approach |
|---|---|---|
| Primary designer | Humans | Humans working with AI |
| Exploring alternatives | Adjusting one point at a time | Exploring multiple alternatives simultaneously |
| Human role | Defining and adjusting the design | Defining goals and reviewing the design |
How This Concept Changes Engineers’ Working Lives
AI can discover chip designs that a team might overlook, such as selecting an approach suited to workloads on the Apple A19 Pro (3 nm) instead of adjusting the design point by point.
When AI models have different objectives, engineers can ask AI to propose hardware suited to each task while examining results on real devices, such as a 120Hz OLED display or a 48 MP camera system.
Simulations of multiple alternatives can be run more quickly through the same tools. Engineers therefore have more time to compare advantages and disadvantages and eliminate unsuitable designs.
People remain just as important, but their role shifts from manually adjusting every parameter to defining goals, reviewing test results, and deciding which designs are ready for real-world use.
How This Concept Changes Engineers’ Working Lives
AI can discover chip designs that a team might overlook, such as selecting an approach suited to workloads on the Apple A19 Pro (3 nm) instead of adjusting the design point by point.
When AI models have different objectives, engineers can ask AI to propose hardware suited to each task while examining results on real devices, such as a 120Hz OLED display or a 48 MP camera system.
Simulations of multiple alternatives can be run more quickly through the same tools. Engineers therefore have more time to compare advantages and disadvantages and eliminate unsuitable designs.
People remain just as important, but their role shifts from manually adjusting every parameter to defining goals, reviewing test results, and deciding which designs are ready for real-world use.
Other Approaches Bringing AI Closer to Hardware Design
Ricursive Intelligence stands out by having AI search through multiple circuit designs, while human teams retain control over the goals and final decisions. Traditional design is slower, but its reasoning is more straightforward to inspect.
| Factor | Ricursive Intelligence | Traditional human team | AI-assisted EDA | Specialized chips from cloud companies |
|---|---|---|---|---|
| Flexibility | High | Limited by team time | Moderate | Focused on specialized tasks |
| Speed | High | Low | High | High |
| Reliability | Requires multiple levels of review | Steps are easy to explain | Depends on the tool | Depends on usage |
| Human control | Controls goals and approval | Controls every step | Controls tool usage | Controls within platform limits |
Other Approaches Bringing AI Closer to Hardware Design
Ricursive Intelligence stands out by having AI search through multiple circuit designs, while human teams retain control over the goals and final decisions. Traditional design is slower, but its reasoning is more straightforward to inspect.
| Factor | Ricursive Intelligence | Traditional human team | AI-assisted EDA | Specialized chips from cloud companies |
|---|---|---|---|---|
| Flexibility | High | Limited by team time | Moderate | Focused on specialized tasks |
| Speed | High | Low | High | High |
| Reliability | Requires multiple levels of review | Steps are easy to explain | Depends on the tool | Depends on usage |
| Human control | Controls goals and approval | Controls every step | Controls tool usage | Controls within platform limits |
Strengths to Watch and Questions Still Unanswered
AI-assisted hardware design may accelerate experimentation and open the way for new architectures better suited to specialized tasks. However, the results still need to be checked repeatedly to ensure that they are accurate, explainable, and safe enough for real-world use.
Another challenge is that a design that looks good on paper may be difficult to manufacture because of limitations involving materials, costs, and production processes.
Pros
- +Accelerates design experimentation
- +Better adapts hardware to specific workloads
Cons
- −Requires accuracy and safety verification
- −The process may be opaque and difficult to manufacture
Strengths to Watch and Questions Still Unanswered
AI-assisted hardware design may accelerate experimentation and open the way for new architectures better suited to specialized tasks. However, the results still need to be checked repeatedly to ensure that they are accurate, explainable, and safe enough for real-world use.
Another challenge is that a design that looks good on paper may be difficult to manufacture because of limitations involving materials, costs, and production processes.
Pros
- +Accelerates design experimentation
- +Better adapts hardware to specific workloads
Cons
- −Requires accuracy and safety verification
- −The process may be opaque and difficult to manufacture
The Price to Pay Even Without Buying a Single Chip
Having AI design hardware involves more than model usage costs. It also requires infrastructure for training and simulations, as well as people who understand AI, circuitry, and real-world manufacturing processes.
Teams must also spend time verifying accuracy, safety, and production requirements. If a design that looks good in software cannot be built, costs increase through redesigns, new prototypes, and delayed delivery schedules.
The Price to Pay Even Without Buying a Single Chip
Having AI design hardware involves more than model usage costs. It also requires infrastructure for training and simulations, as well as people who understand AI, circuitry, and real-world manufacturing processes.
Teams must also spend time verifying accuracy, safety, and production requirements. If a design that looks good in software cannot be built, costs increase through redesigns, new prototypes, and delayed delivery schedules.
When AI Becomes a Co-Designer, Who Makes the Decisions?
Conversations with Anna Goldie and Azalia Mirhoseini illustrate that AI is not merely helping find answers. It is beginning to propose computer architecture designs that humans might overlook. The challenge, therefore, is not only how well AI can design, but also who defines the problem and the criteria for deciding whether a design is suitable for real-world use.
The future may not mean AI replacing engineers, but rather changing their role from drawing every line themselves to defining goals, reviewing results, and taking responsibility for the decisions made by automated design systems.
When AI Becomes a Co-Designer, Who Makes the Decisions?
Conversations with Anna Goldie and Azalia Mirhoseini illustrate that AI is not merely helping find answers. It is beginning to propose computer architecture designs that humans might overlook. The challenge, therefore, is not only how well AI can design, but also who defines the problem and the criteria for deciding whether a design is suitable for real-world use.
The future may not mean AI replacing engineers, but rather changing their role from drawing every line themselves to defining goals, reviewing results, and taking responsibility for the decisions made by automated design systems.
When AI Does More Than Use Chips—it Starts Inventing Them
At TechCrunch Disrupt, Anna Goldie and Azalia Mirhoseini offered another way to view AI. It is not merely a chip user; it is beginning to help discover chip design approaches on its own.
The turning point is that hardware design may become a collaboration between people and AI. Engineers define the goals and constraints, while AI works through complex alternatives before handing them back to humans for review and decision-making.
When AI Does More Than Use Chips—it Starts Inventing Them
At TechCrunch Disrupt, Anna Goldie and Azalia Mirhoseini offered another way to view AI. It is not merely a chip user; it is beginning to help discover chip design approaches on its own.
The turning point is that hardware design may become a collaboration between people and AI. Engineers define the goals and constraints, while AI works through complex alternatives before handing them back to humans for review and decision-making.
When Chip Design Can No Longer Keep Up with AI’s Speed
Engineering teams are facing new types of computational challenges that are changing faster than conventional hardware can keep up. A chip that once suited one task may no longer meet requirements when AI needs to process a different type of workload soon afterward.
This was the question Anna Goldie and Azalia Mirhoseini raised at TechCrunch Disrupt 2026: if AI can discover hardware structures suited to a task on its own, how much faster can the design cycle become? And will engineers shift from trying every possible option to defining goals, reviewing results, and selecting the approaches that work in practice?
When Chip Design Can No Longer Keep Up with AI’s Speed
Engineering teams are facing new types of computational challenges that are changing faster than conventional hardware can keep up. A chip that once suited one task may no longer meet requirements when AI needs to process a different type of workload soon afterward.
This was the question Anna Goldie and Azalia Mirhoseini raised at TechCrunch Disrupt 2026: if AI can discover hardware structures suited to a task on its own, how much faster can the design cycle become? And will engineers shift from trying every possible option to defining goals, reviewing results, and selecting the approaches that work in practice?
Where Ricursive Intelligence Fits in the AI and Chip Ecosystem
Ricursive Intelligence sits between AI model developers and chip design teams. The company does not create models for direct user interaction; instead, it uses AI to find hardware structures suited to different types of workloads.
Viewed as a system, chip manufacturers build the platform, while specialized computer-system design teams determine how to use its resources most efficiently. Ricursive therefore connects these two sides, allowing AI to propose alternatives while engineers review their speed, cost, and manufacturability.
Where Ricursive Intelligence Fits in the AI and Chip Ecosystem
Ricursive Intelligence sits between AI model developers and chip design teams. The company does not create models for direct user interaction; instead, it uses AI to find hardware structures suited to different types of workloads.
Viewed as a system, chip manufacturers build the platform, while specialized computer-system design teams determine how to use its resources most efficiently. Ricursive therefore connects these two sides, allowing AI to propose alternatives while engineers review their speed, cost, and manufacturability.
From Humans Defining Everything to AI Joining the Design Process
The traditional approach has humans establish the architecture, choose the algorithms, and fine-tune the design manually. The process therefore follows the options visible to the team. The new approach allows AI to explore hardware structures and a large number of alternatives simultaneously before sending them to engineers to review for speed, cost, and manufacturability.
| Factor | Traditional approach | AI-assisted design |
|---|---|---|
| Architecture defined by | Primarily humans | Humans working with AI |
| Exploring alternatives | Refining a predetermined approach | Exploring numerous structures and algorithms |
| Final decision | Engineering team | Engineering team reviews AI proposals |
The turning point is therefore not allowing AI to make every decision, but helping teams see more options sooner and then selecting designs that can actually be built.
From Humans Defining Everything to AI Joining the Design Process
The traditional approach has humans establish the architecture, choose the algorithms, and fine-tune the design manually. The process therefore follows the options visible to the team. The new approach allows AI to explore hardware structures and a large number of alternatives simultaneously before sending them to engineers to review for speed, cost, and manufacturability.
| Factor | Traditional approach | AI-assisted design |
|---|---|---|
| Architecture defined by | Primarily humans | Humans working with AI |
| Exploring alternatives | Refining a predetermined approach | Exploring numerous structures and algorithms |
| Final decision | Engineering team | Engineering team reviews AI proposals |
The turning point is therefore not allowing AI to make every decision, but helping teams see more options sooner and then selecting designs that can actually be built.
How This Concept Changes Engineers’ Working Lives
When designing chips for AI workloads, engineers no longer have to begin with only a handful of options. AI can discover structures that humans may overlook and propose designs for the team to review.
Different tasks may require different hardware. For example, a model that needs fast response times might be paired with the Apple A19 Pro (3 nm) and a 120Hz OLED display so that processing and rendering work in harmony.
Instead of testing each design one at a time, simulation tools can filter a large number of alternatives first. Teams can then spend more time on designs with a greater chance of being built in practice.
Ultimately, AI, simulation tools, and engineers must work as one continuous loop. AI proposes, the team checks the constraints, and the results are fed back into the next design iteration.
How This Concept Changes Engineers’ Working Lives
When designing chips for AI workloads, engineers no longer have to begin with only a handful of options. AI can discover structures that humans may overlook and propose designs for the team to review.
Different tasks may require different hardware. For example, a model that needs fast response times might be paired with the Apple A19 Pro (3 nm) and a 120Hz OLED display so that processing and rendering work in harmony.
Instead of testing each design one at a time, simulation tools can filter a large number of alternatives first. Teams can then spend more time on designs with a greater chance of being built in practice.
Ultimately, AI, simulation tools, and engineers must work as one continuous loop. AI proposes, the team checks the constraints, and the results are fed back into the next design iteration.
Other Approaches Bringing AI Closer to Hardware Design
| Factor | Ricursive Intelligence | Traditional human team | AI-assisted EDA | Specialized chips from cloud companies |
|---|---|---|---|---|
| Flexibility | High, with broad room to adjust approaches | Limited by team time | High within the tool’s framework | Focused on specialized tasks |
| Speed | Fast because it tests multiple approaches | Slower when designs require multiple revisions | Faster for repetitive tasks | Fast when the problem is clearly defined |
| Reliability | Requires review by engineers | Relies on team experience | Requires reviewing AI results | Controlled through the company’s systems |
| Human control | Collaborative design and oversight | Direct control | Still controls the key steps | Depends on the provider |
Ricursive stands out by giving AI room to propose new designs. AI-assisted EDA is suited to teams that want to accelerate existing work, while human-led design still has an advantage in detailed decision-making and constraint checking.
Other Approaches Bringing AI Closer to Hardware Design
| Factor | Ricursive Intelligence | Traditional human team | AI-assisted EDA | Specialized chips from cloud companies |
|---|---|---|---|---|
| Flexibility | High, with broad room to adjust approaches | Limited by team time | High within the tool’s framework | Focused on specialized tasks |
| Speed | Fast because it tests multiple approaches | Slower when designs require multiple revisions | Faster for repetitive tasks | Fast when the problem is clearly defined |
| Reliability | Requires review by engineers | Relies on team experience | Requires reviewing AI results | Controlled through the company’s systems |
| Human control | Collaborative design and oversight | Direct control | Still controls the key steps | Depends on the provider |
Ricursive stands out by giving AI room to propose new designs. AI-assisted EDA is suited to teams that want to accelerate existing work, while human-led design still has an advantage in detailed decision-making and constraint checking.
Strengths to Watch and Questions Still Unanswered
AI can accelerate experimentation and may open the way for new architectures beyond human expectations by adapting hardware more closely to each type of workload.
However, it still needs to be proven that the results are accurate, traceable, and safe enough for real-world manufacturing. Most importantly, a design that looks good on paper may be difficult or expensive to manufacture.
Pros
- +Accelerates design experimentation
- +Creates room for new architectures
- +Better adapts hardware to specific workloads
Cons
- −Accuracy is difficult to verify
- −The process is not yet transparent enough
- −Safety and real-world manufacturability remain open questions
Strengths to Watch and Questions Still Unanswered
AI can accelerate experimentation and may open the way for new architectures beyond human expectations by adapting hardware more closely to each type of workload.
However, it still needs to be proven that the results are accurate, traceable, and safe enough for real-world manufacturing. Most importantly, a design that looks good on paper may be difficult or expensive to manufacture.
Pros
- +Accelerates design experimentation
- +Creates room for new architectures
- +Better adapts hardware to specific workloads
Cons
- −Accuracy is difficult to verify
- −The process is not yet transparent enough
- −Safety and real-world manufacturability remain open questions
The Price to Pay Even Without Buying a Single Chip
The cost is not limited to running AI. It also includes infrastructure for training models and running simulations. Teams need people who understand software, circuitry, and real-world manufacturing processes to filter designs that look good on screen.
Review time is another major cost, because every design must be checked for accuracy, safety, and performance under real-world conditions. If an AI-generated design cannot be manufactured, a company may lose time, prototype costs, and business opportunities without obtaining a usable chip.
The Price to Pay Even Without Buying a Single Chip
The cost is not limited to running AI. It also includes infrastructure for training models and running simulations. Teams need people who understand software, circuitry, and real-world manufacturing processes to filter designs that look good on screen.
Review time is another major cost, because every design must be checked for accuracy, safety, and performance under real-world conditions. If an AI-generated design cannot be manufactured, a company may lose time, prototype costs, and business opportunities without obtaining a usable chip.
When AI Becomes a Co-Designer, Who Makes the Decisions?
AI may explore chip alternatives more quickly, but the answer does not end with a design on a screen. Engineers still need to ask whether the design suits real-world workloads, whether its cost is acceptable, and whether it can actually be manufactured.
Consider the Apple A19 Pro chip, built using a 3nm process. A design that appears powerful may not be enough if it must work alongside a 120Hz OLED display, a 48 MP camera, and a real device battery. Every component must remain balanced rather than optimizing a single number.
I think the future may not mean AI replacing engineers, but rather changing their role from drawing every line themselves to defining goals, reviewing results, and taking responsibility for the decisions made by automated design systems.
When AI Becomes a Co-Designer, Who Makes the Decisions?
AI may explore chip alternatives more quickly, but the answer does not end with a design on a screen. Engineers still need to ask whether the design suits real-world workloads, whether its cost is acceptable, and whether it can actually be manufactured.
Consider the Apple A19 Pro chip, built using a 3nm process. A design that appears powerful may not be enough if it must work alongside a 120Hz OLED display, a 48 MP camera, and a real device battery. Every component must remain balanced rather than optimizing a single number.
I think the future may not mean AI replacing engineers, but rather changing their role from drawing every line themselves to defining goals, reviewing results, and taking responsibility for the decisions made by automated design systems. This chapter explores how AI may evolve from a tool that assists with hardware design into a co-inventor of computer architectures, using familiar examples such as the Apple A19 Pro, a 3 nm chip, and 12 GB of RAM in the iPhone 17 Pro Max.
The key issue is not simply that AI can design chips faster, but the limitations surrounding accuracy, safety, and result verification. If a system makes a mistake at the architectural level, the effects could extend to real devices, from performance to power consumption. This article therefore examines the opportunities, risks, and future of the chip industry from all angles.
This chapter explores how AI may evolve from a tool that assists with hardware design into a co-inventor of computer architectures, using familiar examples such as the Apple A19 Pro, a 3 nm chip, and 12 GB of RAM in the iPhone 17 Pro Max.
The key issue is not simply that AI can design chips faster, but the limitations surrounding accuracy, safety, and result verification. If a system makes a mistake at the architectural level, the effects could extend to real devices, from performance to power consumption. This article therefore examines the opportunities, risks, and future of the chip industry from all angles.
When AI Does More Than Use Chips—it Starts Inventing Them
This reflects a major turning point in the industry, as AI is moving from being a coding assistant to becoming a tool that can help devise chip structures itself. The work of Anna Goldie and Azalia Mirhoseini suggests that AI may help uncover options humans overlook, although the results still require serious verification.
Consider the Apple A19 Pro (3 nm) in the iPhone 17 Pro Max: the story behind a smaller chip is not just about speed, but also heat, battery life, and stability. If AI makes a design mistake, the impact can immediately reach real-world use.
When AI Does More Than Use Chips—it Starts Inventing Them
This reflects a major turning point in the industry, as AI is moving from being a coding assistant to becoming a tool that can help devise chip structures itself. The work of Anna Goldie and Azalia Mirhoseini suggests that AI may help uncover options humans overlook, although the results still require serious verification.
Consider the Apple A19 Pro (3 nm) in the iPhone 17 Pro Max: the story behind a smaller chip is not just about speed, but also heat, battery life, and stability. If AI makes a design mistake, the impact can immediately reach real-world use.
When Chip Design Can No Longer Keep Up with AI’s Speed
Engineering teams are facing challenges that are changing increasingly quickly, including computational patterns, model sizes, and the tasks AI must handle. Conventional hardware may therefore fail to keep up with new demands, even if it still performs well for its original purpose.
The important question is: if AI can help discover chip structures and test alternatives on its own, how much faster can the design cycle become? The turning point is not about letting AI make every decision for humans, but about opening up new possibilities and having engineers verify whether they are feasible, stable, and suitable for the intended use.
When Chip Design Can No Longer Keep Up with AI’s Speed
Engineering teams are facing challenges that are changing increasingly quickly, including computational patterns, model sizes, and the tasks AI must handle. Conventional hardware may therefore fail to keep up with new demands, even if it still performs well for its original purpose.
The important question is: if AI can help discover chip structures and test alternatives on its own, how much faster can the design cycle become? The turning point is not about letting AI make every decision for humans, but about opening up new possibilities and having engineers verify whether they are feasible, stable, and suitable for the intended use.
Where Ricursive Intelligence Fits in the AI and Chip Ecosystem
Ricursive Intelligence sits between AI model developers and teams designing specialized computer systems. It uses AI to explore hardware designs suited to real-world tasks rather than replacing chip manufacturers responsible for factory production.
Under the vision of Anna Goldie and Azalia Mirhoseini, the company functions as a bridge between software and hardware. AI models need chips suited to their workloads, while chip manufacturers need designs that can actually be built and support long-term use.
Where Ricursive Intelligence Fits in the AI and Chip Ecosystem
Ricursive Intelligence sits between AI model developers and teams designing specialized computer systems. It uses AI to explore hardware designs suited to real-world tasks rather than replacing chip manufacturers responsible for factory production.
Under the vision of Anna Goldie and Azalia Mirhoseini, the company functions as a bridge between software and hardware. AI models need chips suited to their workloads, while chip manufacturers need designs that can actually be built and support long-term use.
From Humans Defining Everything to AI Joining the Design Process
The traditional approach has humans establish the architecture and then manually fine-tune it point by point. This works well when the scope is clearly defined, but becomes time-consuming as the number of options increases.
The new approach allows AI to explore hardware structures, algorithms, and multiple alternatives simultaneously. Humans still define the goals, assess feasibility, and make the final decisions.
| Factor | Traditional approach | New approach |
|---|---|---|
| Primary designer | Humans | Humans working with AI |
| Exploring alternatives | Adjusting one point at a time | Exploring multiple alternatives simultaneously |
| Human role | Defining and adjusting the design | Defining goals and reviewing the design |
From Humans Defining Everything to AI Joining the Design Process
The traditional approach has humans establish the architecture and then manually fine-tune it point by point. This works well when the scope is clearly defined, but becomes time-consuming as the number of options increases.
The new approach allows AI to explore hardware structures, algorithms, and multiple alternatives simultaneously. Humans still define the goals, assess feasibility, and make the final decisions.
| Factor | Traditional approach | New approach |
|---|---|---|
| Primary designer | Humans | Humans working with AI |
| Exploring alternatives | Adjusting one point at a time | Exploring multiple alternatives simultaneously |
| Human role | Defining and adjusting the design | Defining goals and reviewing the design |
How This Concept Changes Engineers’ Working Lives
AI can discover chip designs that a team might overlook, such as selecting an approach suited to workloads on the Apple A19 Pro (3 nm) instead of adjusting the design point by point.
When AI models have different objectives, engineers can ask AI to propose hardware suited to each task while examining results on real devices, such as a 120Hz OLED display or a 48 MP camera system.
Simulations of multiple alternatives can be run more quickly through the same tools. Engineers therefore have more time to compare advantages and disadvantages and eliminate unsuitable designs.
People remain just as important, but their role shifts from manually adjusting every parameter to defining goals, reviewing test results, and deciding which designs are ready for real-world use.
How This Concept Changes Engineers’ Working Lives
AI can discover chip designs that a team might overlook, such as selecting an approach suited to workloads on the Apple A19 Pro (3 nm) instead of adjusting the design point by point.
When AI models have different objectives, engineers can ask AI to propose hardware suited to each task while examining results on real devices, such as a 120Hz OLED display or a 48 MP camera system.
Simulations of multiple alternatives can be run more quickly through the same tools. Engineers therefore have more time to compare advantages and disadvantages and eliminate unsuitable designs.
People remain just as important, but their role shifts from manually adjusting every parameter to defining goals, reviewing test results, and deciding which designs are ready for real-world use.
Other Approaches Bringing AI Closer to Hardware Design
Ricursive Intelligence stands out by having AI search through multiple circuit designs, while human teams retain control over the goals and final decisions. Traditional design is slower, but its reasoning is more straightforward to inspect.
| Factor | Ricursive Intelligence | Traditional human team | AI-assisted EDA | Specialized chips from cloud companies |
|---|---|---|---|---|
| Flexibility | High | Limited by team time | Moderate | Focused on specialized tasks |
| Speed | High | Low | High | High |
| Reliability | Requires multiple levels of review | Steps are easy to explain | Depends on the tool | Depends on usage |
| Human control | Controls goals and approval | Controls every step | Controls tool usage | Controls within platform limits |
Other Approaches Bringing AI Closer to Hardware Design
Ricursive Intelligence stands out by having AI search through multiple circuit designs, while human teams retain control over the goals and final decisions. Traditional design is slower, but its reasoning is more straightforward to inspect.
| Factor | Ricursive Intelligence | Traditional human team | AI-assisted EDA | Specialized chips from cloud companies |
|---|---|---|---|---|
| Flexibility | High | Limited by team time | Moderate | Focused on specialized tasks |
| Speed | High | Low | High | High |
| Reliability | Requires multiple levels of review | Steps are easy to explain | Depends on the tool | Depends on usage |
| Human control | Controls goals and approval | Controls every step | Controls tool usage | Controls within platform limits |
Strengths to Watch and Questions Still Unanswered
AI-assisted hardware design may accelerate experimentation and open the way for new architectures better suited to specialized tasks. However, the results still need to be checked repeatedly to ensure that they are accurate, explainable, and safe enough for real-world use.
Another challenge is that a design that looks good on paper may be difficult to manufacture because of limitations involving materials, costs, and production processes.
Pros
- +Accelerates design experimentation
- +Better adapts hardware to specific workloads
Cons
- −Requires accuracy and safety verification
- −The process may be opaque and difficult to manufacture
Strengths to Watch and Questions Still Unanswered
AI-assisted hardware design may accelerate experimentation and open the way for new architectures better suited to specialized tasks. However, the results still need to be checked repeatedly to ensure that they are accurate, explainable, and safe enough for real-world use.
Another challenge is that a design that looks good on paper may be difficult to manufacture because of limitations involving materials, costs, and production processes.
Pros
- +Accelerates design experimentation
- +Better adapts hardware to specific workloads
Cons
- −Requires accuracy and safety verification
- −The process may be opaque and difficult to manufacture
The Price to Pay Even Without Buying a Single Chip
Having AI design hardware involves more than model usage costs. It also requires infrastructure for training and simulations, as well as people who understand AI, circuitry, and real-world manufacturing processes.
Teams must also spend time verifying accuracy, safety, and production requirements. If a design that looks good in software cannot be built, costs increase through redesigns, new prototypes, and delayed delivery schedules.
The Price to Pay Even Without Buying a Single Chip
Having AI design hardware involves more than model usage costs. It also requires infrastructure for training and simulations, as well as people who understand AI, circuitry, and real-world manufacturing processes.
Teams must also spend time verifying accuracy, safety, and production requirements. If a design that looks good in software cannot be built, costs increase through redesigns, new prototypes, and delayed delivery schedules.
When AI Becomes a Co-Designer, Who Makes the Decisions?
Conversations with Anna Goldie and Azalia Mirhoseini illustrate that AI is not merely helping find answers. It is beginning to propose computer architecture designs that humans might overlook. The challenge, therefore, is not only how well AI can design, but also who defines the problem and the criteria for deciding whether a design is suitable for real-world use.
The future may not mean AI replacing engineers, but rather changing their role from drawing every line themselves to defining goals, reviewing results, and taking responsibility for the decisions made by automated design systems.
When AI Becomes a Co-Designer, Who Makes the Decisions?
Conversations with Anna Goldie and Azalia Mirhoseini illustrate that AI is not merely helping find answers. It is beginning to propose computer architecture designs that humans might overlook. The challenge, therefore, is not only how well AI can design, but also who defines the problem and the criteria for deciding whether a design is suitable for real-world use.
The future may not mean AI replacing engineers, but rather changing their role from drawing every line themselves to defining goals, reviewing results, and taking responsibility for the decisions made by automated design systems.
When AI Does More Than Use Chips—it Starts Inventing Them
At TechCrunch Disrupt, Anna Goldie and Azalia Mirhoseini offered another way to view AI. It is not merely a chip user; it is beginning to help discover chip design approaches on its own.
The turning point is that hardware design may become a collaboration between people and AI. Engineers define the goals and constraints, while AI works through complex alternatives before handing them back to humans for review and decision-making.
When AI Does More Than Use Chips—it Starts Inventing Them
At TechCrunch Disrupt, Anna Goldie and Azalia Mirhoseini offered another way to view AI. It is not merely a chip user; it is beginning to help discover chip design approaches on its own.
The turning point is that hardware design may become a collaboration between people and AI. Engineers define the goals and constraints, while AI works through complex alternatives before handing them back to humans for review and decision-making.
When Chip Design Can No Longer Keep Up with AI’s Speed
Engineering teams are facing new types of computational challenges that are changing faster than conventional hardware can keep up. A chip that once suited one task may no longer meet requirements when AI needs to process a different type of workload soon afterward.
This was the question Anna Goldie and Azalia Mirhoseini raised at TechCrunch Disrupt 2026: if AI can discover hardware structures suited to a task on its own, how much faster can the design cycle become? And will engineers shift from trying every possible option to defining goals, reviewing results, and selecting the approaches that work in practice?
When Chip Design Can No Longer Keep Up with AI’s Speed
Engineering teams are facing new types of computational challenges that are changing faster than conventional hardware can keep up. A chip that once suited one task may no longer meet requirements when AI needs to process a different type of workload soon afterward.
This was the question Anna Goldie and Azalia Mirhoseini raised at TechCrunch Disrupt 2026: if AI can discover hardware structures suited to a task on its own, how much faster can the design cycle become? And will engineers shift from trying every possible option to defining goals, reviewing results, and selecting the approaches that work in practice?
Where Ricursive Intelligence Fits in the AI and Chip Ecosystem
Ricursive Intelligence sits between AI model developers and chip design teams. The company does not create models for direct user interaction; instead, it uses AI to find hardware structures suited to different types of workloads.
Viewed as a system, chip manufacturers build the platform, while specialized computer-system design teams determine how to use its resources most efficiently. Ricursive therefore connects these two sides, allowing AI to propose alternatives while engineers review their speed, cost, and manufacturability.
Where Ricursive Intelligence Fits in the AI and Chip Ecosystem
Ricursive Intelligence sits between AI model developers and chip design teams. The company does not create models for direct user interaction; instead, it uses AI to find hardware structures suited to different types of workloads.
Viewed as a system, chip manufacturers build the platform, while specialized computer-system design teams determine how to use its resources most efficiently. Ricursive therefore connects these two sides, allowing AI to propose alternatives while engineers review their speed, cost, and manufacturability.
From Humans Defining Everything to AI Joining the Design Process
The traditional approach has humans establish the architecture, choose the algorithms, and fine-tune the design manually. The process therefore follows the options visible to the team. The new approach allows AI to explore hardware structures and a large number of alternatives simultaneously before sending them to engineers to review for speed, cost, and manufacturability.
| Factor | Traditional approach | AI-assisted design |
|---|---|---|
| Architecture defined by | Primarily humans | Humans working with AI |
| Exploring alternatives | Refining a predetermined approach | Exploring numerous structures and algorithms |
| Final decision | Engineering team | Engineering team reviews AI proposals |
The turning point is therefore not allowing AI to make every decision, but helping teams see more options sooner and then selecting designs that can actually be built.
From Humans Defining Everything to AI Joining the Design Process
The traditional approach has humans establish the architecture, choose the algorithms, and fine-tune the design manually. The process therefore follows the options visible to the team. The new approach allows AI to explore hardware structures and a large number of alternatives simultaneously before sending them to engineers to review for speed, cost, and manufacturability.
| Factor | Traditional approach | AI-assisted design |
|---|---|---|
| Architecture defined by | Primarily humans | Humans working with AI |
| Exploring alternatives | Refining a predetermined approach | Exploring numerous structures and algorithms |
| Final decision | Engineering team | Engineering team reviews AI proposals |
The turning point is therefore not allowing AI to make every decision, but helping teams see more options sooner and then selecting designs that can actually be built.
How This Concept Changes Engineers’ Working Lives
When designing chips for AI workloads, engineers no longer have to begin with only a handful of options. AI can discover structures that humans may overlook and propose designs for the team to review.
Different tasks may require different hardware. For example, a model that needs fast response times might be paired with the Apple A19 Pro (3 nm) and a 120Hz OLED display so that processing and rendering work in harmony.
Instead of testing each design one at a time, simulation tools can filter a large number of alternatives first. Teams can then spend more time on designs with a greater chance of being built in practice.
Ultimately, AI, simulation tools, and engineers must work as one continuous loop. AI proposes, the team checks the constraints, and the results are fed back into the next design iteration.
How This Concept Changes Engineers’ Working Lives
When designing chips for AI workloads, engineers no longer have to begin with only a handful of options. AI can discover structures that humans may overlook and propose designs for the team to review.
Different tasks may require different hardware. For example, a model that needs fast response times might be paired with the Apple A19 Pro (3 nm) and a 120Hz OLED display so that processing and rendering work in harmony.
Instead of testing each design one at a time, simulation tools can filter a large number of alternatives first. Teams can then spend more time on designs with a greater chance of being built in practice.
Ultimately, AI, simulation tools, and engineers must work as one continuous loop. AI proposes, the team checks the constraints, and the results are fed back into the next design iteration.
Other Approaches Bringing AI Closer to Hardware Design
| Factor | Ricursive Intelligence | Traditional human team | AI-assisted EDA | Specialized chips from cloud companies |
|---|---|---|---|---|
| Flexibility | High, with broad room to adjust approaches | Limited by team time | High within the tool’s framework | Focused on specialized tasks |
| Speed | Fast because it tests multiple approaches | Slower when designs require multiple revisions | Faster for repetitive tasks | Fast when the problem is clearly defined |
| Reliability | Requires review by engineers | Relies on team experience | Requires reviewing AI results | Controlled through the company’s systems |
| Human control | Collaborative design and oversight | Direct control | Still controls the key steps | Depends on the provider |
Ricursive stands out by giving AI room to propose new designs. AI-assisted EDA is suited to teams that want to accelerate existing work, while human-led design still has an advantage in detailed decision-making and constraint checking.
Other Approaches Bringing AI Closer to Hardware Design
| Factor | Ricursive Intelligence | Traditional human team | AI-assisted EDA | Specialized chips from cloud companies |
|---|---|---|---|---|
| Flexibility | High, with broad room to adjust approaches | Limited by team time | High within the tool’s framework | Focused on specialized tasks |
| Speed | Fast because it tests multiple approaches | Slower when designs require multiple revisions | Faster for repetitive tasks | Fast when the problem is clearly defined |
| Reliability | Requires review by engineers | Relies on team experience | Requires reviewing AI results | Controlled through the company’s systems |
| Human control | Collaborative design and oversight | Direct control | Still controls the key steps | Depends on the provider |
Ricursive stands out by giving AI room to propose new designs. AI-assisted EDA is suited to teams that want to accelerate existing work, while human-led design still has an advantage in detailed decision-making and constraint checking.
Strengths to Watch and Questions Still Unanswered
AI can accelerate experimentation and may open the way for new architectures beyond human expectations by adapting hardware more closely to each type of workload.
However, it still needs to be proven that the results are accurate, traceable, and safe enough for real-world manufacturing. Most importantly, a design that looks good on paper may be difficult or expensive to manufacture.
Pros
- +Accelerates design experimentation
- +Creates room for new architectures
- +Better adapts hardware to specific workloads
Cons
- −Accuracy is difficult to verify
- −The process is not yet transparent enough
- −Safety and real-world manufacturability remain open questions
Strengths to Watch and Questions Still Unanswered
AI can accelerate experimentation and may open the way for new architectures beyond human expectations by adapting hardware more closely to each type of workload.
However, it still needs to be proven that the results are accurate, traceable, and safe enough for real-world manufacturing. Most importantly, a design that looks good on paper may be difficult or expensive to manufacture.
Pros
- +Accelerates design experimentation
- +Creates room for new architectures
- +Better adapts hardware to specific workloads
Cons
- −Accuracy is difficult to verify
- −The process is not yet transparent enough
- −Safety and real-world manufacturability remain open questions
The Price to Pay Even Without Buying a Single Chip
The cost is not limited to running AI. It also includes infrastructure for training models and running simulations. Teams need people who understand software, circuitry, and real-world manufacturing processes to filter designs that look good on screen.
Review time is another major cost, because every design must be checked for accuracy, safety, and performance under real-world conditions. If an AI-generated design cannot be manufactured, a company may lose time, prototype costs, and business opportunities without obtaining a usable chip.
The Price to Pay Even Without Buying a Single Chip
The cost is not limited to running AI. It also includes infrastructure for training models and running simulations. Teams need people who understand software, circuitry, and real-world manufacturing processes to filter designs that look good on screen.
Review time is another major cost, because every design must be checked for accuracy, safety, and performance under real-world conditions. If an AI-generated design cannot be manufactured, a company may lose time, prototype costs, and business opportunities without obtaining a usable chip.
When AI Becomes a Co-Designer, Who Makes the Decisions?
AI may explore chip alternatives more quickly, but the answer does not end with a design on a screen. Engineers still need to ask whether the design suits real-world workloads, whether its cost is acceptable, and whether it can actually be manufactured.
Consider the Apple A19 Pro chip, built using a 3nm process. A design that appears powerful may not be enough if it must work alongside a 120Hz OLED display, a 48 MP camera, and a real device battery. Every component must remain balanced rather than optimizing a single number.
I think the future may not mean AI replacing engineers, but rather changing their role from drawing every line themselves to defining goals, reviewing results, and taking responsibility for the decisions made by automated design systems.
When AI Becomes a Co-Designer, Who Makes the Decisions?
AI may explore chip alternatives more quickly, but the answer does not end with a design on a screen. Engineers still need to ask whether the design suits real-world workloads, whether its cost is acceptable, and whether it can actually be manufactured.
Consider the Apple A19 Pro chip, built using a 3nm process. A design that appears powerful may not be enough if it must work alongside a 120Hz OLED display, a 48 MP camera, and a real device battery. Every component must remain balanced rather than optimizing a single number.
I think the future may not mean AI replacing engineers, but rather changing their role from drawing every line themselves to defining goals, reviewing results, and taking responsibility for the decisions made by automated design systems.