Will AI Undertake Most Work by 2035? A Look into DeepMind's Ambitious Timeline

Recent predictions from leaders at Google DeepMind have sparked renewed discussions about the timeline for AI taking over the majority of work currently done by humans. Several key figures within the lab believe that by the early to mid-2030s, AI agents could manage most knowledge work tasks with little supervision. This bold assertion warrants a thoughtful and nuanced exploration.

Quick Overview

DeepMind leaders have predicted that human-level AI could appear before 2030, with widespread labor automation potentially occurring by 2035. These are speculations rather than guaranteed outcomes.
While advancements in multimodal models, tool usage, and AI agents fuel optimism, significant challenges remain, including issues related to data, energy consumption, reliability, robotics, and governance.
Initially, expect AI to enhance existing jobs, followed by the automation of more routine tasks. Complete job replacement across occupations by 2035 is uncertain.
Organizations can begin preparing now by experimenting with AI agents, investing in data quality, enhancing team skills, and developing strong governance structures for AI.

What Did DeepMind Actually State?

There is no formal declaration from DeepMind indicating that AI will take over all work by 2035. However, various senior figures have suggested optimistic timelines for achieving human-level or broadly capable AI. Shane Legg, a co-founder and Chief AGI Scientist at DeepMind, has estimated a 50% chance of reaching human-level AGI by 2028, expecting acceleration in capabilities in the early 2030s. He has reiterated these timelines in multiple discussions over the past few years. Reports of Legg’s predictions in late 2023 and 2024 emphasize a genuine possibility that AI could match or exceed human cognitive performance over the next decade (Business Insider).

Demis Hassabis, CEO of Google DeepMind, has similarly highlighted the rapid advancements in creating general-purpose systems. He pointed out developments like multimodal agents capable of real-time perception and action. Google’s Project Astra, introduced in 2024, is one initiative aimed at building more effective AI assistants (Google DeepMind). While Hassabis has raised concerns about safety risks and stressed the need for responsible development, his remarks suggest that significant advancements could emerge between 2025 and 2035.

In summary, the bold statement about AI performing most work by 2035 is not an official promise but reflects educated yet uncertain forecasts from key players engaged in developing cutting-edge technologies.

Why Experts Consider 2030-2035 Realistic

A variety of technological trends support these ambitious forecasts:

1) Scaling Models and Tool Usage

Cutting-edge models are advancing rapidly as researchers increase the size of datasets, parameters, and compute capabilities. Models are also improving their ability to use tools, making them active problem solvers rather than static responders. This evolution from simple text generation to tool-assisted reasoning and action represents a crucial shift toward the next generation of AI agents.

2) Multimodal Agents That Perceive and Act

Initiatives like Google’s Project Astra showcase real-time comprehension of voice, vision, and text, allowing for immediate analysis of scenes, documents, and interfaces (Google DeepMind). Similarly, OpenAI’s GPT-4o integrates speech, vision, and text to create richer interactive experiences (OpenAI). The push for multimodal capabilities is significant because most real-world tasks involve a blend of activities requiring perceptive interaction with software.

3) Accelerated Coding Support

Developer tools like GitHub Copilot have demonstrated measurable productivity enhancements in controlled environments, enabling faster task completion with fewer interruptions (GitHub Research). As coding agents evolve, they can automate increasing portions of the software development process, which underpins many modern businesses.

4) AI in Scientific Discovery

AI systems are already facilitating significant advancements in scientific research. For instance, AlphaFold 3, launched in 2024, improved accurate protein structure predictions, demonstrating AI’s potential to drive breakthroughs in drug discovery and biology (Nature).

5) Substantial Economic Potential

Research from McKinsey indicates that generative AI could contribute between $2.6 trillion and $4.4 trillion to annual global productivity across various sectors, including marketing, software development, customer operations, and R&D (McKinsey). Other analyses, such as those by Goldman Sachs, project long-term GDP increases due to widespread AI utilization.

Potential Challenges: Why 2035 Might Be Overly Optimistic

While there’s a bullish outlook, several constraints could impede progress:

Data Limitations and Quality

Advanced models have largely utilized top-quality public text and code. Experts caution that we may face a shortage of high-quality data for training larger models, which could restrict performance improvements or force reliance on synthetic data, whose efficacy and risks are still being evaluated (Villalobos et al., 2022).

Energy and Compute Constraints

Running advanced models requires significant computational power and energy. The International Energy Agency foresees rapid rises in power consumption by data centers due to AI, which could have important implications for sustainability and operational costs (IEA). Supply chain issues related to advanced chips and the construction of data centers could also limit scalability in the short term.

Reliability, Safety, and Evaluation

Despite promising demonstrations, models sometimes misinterpret instructions or produce inaccurate results. Developing reliable agents that can navigate complex tasks autonomously necessitates robust evaluation, monitoring, and safety considerations, as highlighted in industry initiatives like Google’s Frontier Safety Framework and Anthropic’s Responsible Scaling proposals (Google) (Anthropic).

Physical Work and Robotics

Many jobs involve physical tasks, and while progress is being made in robotics, translating general AI capabilities into effective manipulation and autonomy in real-world settings remains a challenge. Early initiatives like Google’s RT-2 indicate how vision-language models can steer robots, yet widespread reliable robotics is still on the horizon (Google Research).

Regulatory and Social Adaptation

Governments are moving swiftly regarding AI regulation. The EU AI Act includes a tiered system for high-risk and general-purpose systems, with compliance phases beginning in 2025 (European Commission). The United States has also taken significant steps with a comprehensive Executive Order on AI from 2023 (White House). While essential, such regulations may lengthen deployment timelines in heavily regulated sectors.

Implications of AI Handling Most Work by 2035

There are three primary interpretations of this headline claim:

Task-level automation: AI agents perform most routine tasks across various jobs (e.g., drafting, editing, scheduling, data entry), while humans supervise and manage exceptions.
Workflow-level automation: AI agents manage end-to-end processes (e.g., processing claims, running marketing campaigns), with humans addressing edge cases.
Occupation-level automation: Entire job functions may be executed by AI, encompassing objectives, execution, and both digital and physical environments.

Initial adoption will focus on the first two interpretations, while occupation-level automation will progress more gradually and unevenly, particularly in roles requiring physical presence, nuanced judgment, or extensive human interaction.

Sector Insights

Knowledge Work: Anticipate AI to draft emails, summarize meetings, analyze data, and create presentations. The role of humans will evolve to one of steering and decision-making.
Customer Operations: AI will manage routine inquiries, triage issues, and customize responses, allowing humans to focus on complex cases and building relationships.
Software Development: From generating boilerplate code to producing tests and documentation, AI will handle repetitive elements, while humans oversee architecture and critical code reviews.
Healthcare: AI will support diagnostics, documentation, prior authorizations, and navigating care, with clinical decisions remaining human-led.
Manufacturing and Logistics: Digital optimization will rapidly evolve, while physical automation will depend on advancements in robotics, sensing technologies, and safety protocols.

Insights on Productivity and Employment

Extensive studies suggest that AI can lead to significant productivity improvements, but effects on employment and wages will differ based on role and region.

Productivity Increase: McKinsey estimates generative AI could generate trillions in annual value, particularly within sectors such as customer operations, marketing, sales, software engineering, and R&D (McKinsey). GitHub studies show developers can code significantly faster with tools like Copilot (GitHub Research).
Job Exposure: The IMF estimates that roughly 40% of global jobs are susceptible to AI, with higher vulnerability in advanced economies. Outcomes can range from job enhancement and higher wages to displacement without adequate retraining opportunities (IMF).
Inequality Threat: Without proactive measures, AI may exacerbate income inequality by enhancing productivity in high-skill roles while undermining low-skill wage growth—especially if automation outpaces the formation of new complementary roles (IMF).

Overall, the near future appears more focused on AI-enhanced work rather than widespread job losses. Embracing new roles that involve managing, evaluating, and collaborating with AI systems will be critical.

Safety, Governance, and Implementation Pace

Society is not passively waiting for AGI. In 2023, 28 nations signed the Bletchley Declaration, committing to collaboratively address risks associated with advanced AI, including issues of misuse and loss of control (UK Government). New regulations like the EU AI Act and the US Executive Order on AI have established expectations for risk assessment, transparency, and security in high-risk applications (European Commission) (White House).

Industry leaders are also tightening safety protocols. Google has introduced a Frontier Safety Framework for monitoring and reducing risks as capabilities increase (Google), while Anthropic has shared a Responsible Scaling Policy based on specific benchmarks and tests (Anthropic). Such frameworks will likely extend the timeline for achieving full autonomy in sensitive sectors while still facilitating trust and adoption in other areas.

Practical Steps: How to Get Ready

Regardless of whether the 2035 timeline comes to fruition earlier or later, the trajectory is clear. Here’s how individuals and organizations can prepare:

For Individuals

Engage with AI tools in daily tasks: summarization, drafting, data analysis, and ideation.
Learn skills in prompt designing, tool integration, and basic data management (SQL, spreadsheets, Python notebooks).
Enhance your judgment skills: critical thinking, ethical reasoning, and verification remain key human competencies.
Develop expertise in specific domains that allow you to oversee and refine AI-generated outputs.

For Teams and Leaders

Pursue small pilot projects for high-frequency workflows (customer inquiries, report writing, quality assurance).
Invest in maintaining quality, well-governed data. The quality of data typically dictates AI return on investment more than the choice of model.
Design frameworks that incorporate human oversight into workflows, allowing clear escalation paths for complex scenarios.
Create an AI governance structure: incorporate security reviews, bias assessments, incident management, and audit mechanisms.
Monitor compute and energy usage as adoption scales; work with IT on sustainable practices.

Projected Timelines: Three Scenarios

1) Rapid Development (Optimistic)

2025-2027: Reliable workflows in office software become commonplace. AI manages most writing, data extraction, and routine analysis. 2028-2030: Agents effectively coordinate multi-step tasks across multiple platforms with high reliability. 2030-2035: Widespread workflow automation occurs in administrative sectors, with some physical domains seeing targeted robotics implementations.

2) Steady Advancements (Moderate Perspective)

While capabilities enhance, challenges in data quality, energy, and evaluations persist. Adoption progresses steadily but varies across sectors. By 2035, many jobs will experience augmentation, and some workflows may achieve full automation, though widespread job replacement remains uncommon.

3) Gradual Transition (Conservative)

Challenges related to reliability, regulatory intricacies, and energy constraints delay widespread deployment. AI retains its status as a powerful assistant, but the shift to fully independent agents in critical workflows is postponed beyond 2035.

Key Takeaways

While DeepMind leaders may anticipate a stunning transformation in the early 2030s, the practicalities of the real world introduce various complexities, including safety, regulation, cultural barriers, infrastructure, and trust issues. Forward-thinking preparation for accelerated enhancement is essential, along with building sustainable capabilities in data management and governance to navigate both the opportunities and challenges ahead.

Frequently Asked Questions

Will AI genuinely replace most jobs by 2035?

It’s uncertain. While many tasks will be automated, and some roles may diminish, the majority of positions are expected to evolve rather than disappear, particularly where human judgment, accountability, and physical presence are crucial.

What breakthroughs are still necessary?

Key advancements needed include reliable long-term planning, effective tool usage, verifiable reasoning, scalable evaluation and monitoring systems, and practical robotics. Additionally, efficient models, robust data strategies, and sustainable computing infrastructure will be critical.

How should companies get started?

Focus on high-volume, low-risk workflows to run pilot projects with human oversight. Emphasize investments in data quality and governance, measure results, and refine systems prior to expanding.

What about energy consumption?

The energy demands tied to AI growth will rise as data centers expand. Expect strong incentives for increasingly efficient processes, specialized hardware, and sourcing renewable energy as adoption grows (IEA).

Will regulation hinder progress?

Regulatory frameworks will inevitably influence deployments, especially in high-risk areas. Thoughtfully crafted regulations can foster trust and expedite adoption in the long run, despite introducing short-term challenges.

Will AI Undertake Most Work by 2035? A Look into DeepMind’s Ambitious Timeline