Graph engineering reframes how AI agents work together

What the Heck is Graph Engineering?

As AI systems move beyond isolated prompts, their usefulness increasingly depends on coordinating models, tools, knowledge, and people.

3 key takeaways
  1. 1Graph engineering treats AI as a coordinated system of agents, tools, knowledge, and humans.
  2. 2The framework marks a shift from optimizing individual prompts toward designing relationships and workflows across components.
  3. 3The episode also previews OpenAI’s delayed GPT-6 Astra release alongside other current AI developments.

Don't miss

Whitmore’s central reframing casts graph engineering as the discipline of connecting agents, tools, knowledge, and humans beyond simple prompting.

The brief

Nathaniel Whitmore introduces graph engineering as a way to coordinate AI agents, tools, knowledge, and humans into systems that can do more than isolated prompts.

The core argument is architectural: as AI capabilities expand, effective systems depend on how components connect, delegate, and share context—not only on the model behind each task.

The episode situates the framework amid current AI news, including OpenAI’s delayed GPT-6 Astra release and other developments shaping expectations for agentic systems.

The standout idea is that prompting is becoming one layer of a larger discipline: engineering the graph of relationships that lets AI systems operate coherently.

What was said on this episode

22 statements · 16 positive · 1 negative · 5 neutral

  1. Nathaniel Whittemoreon AI model safety infrastructurePositive3:48

    Nathaniel predicts substantially greater investment supporting unreleased high-risk models.

    “where I believe there will be a significantly increased investment in the resources to properly support models that won't be able to be released to the public without it.”

    Listen at 3:48

  2. Nathaniel Whittemoreon AI model safety resourcesPositive3:49

    AI labs will significantly increase resources supporting models requiring stronger safeguards.

    “there will be a significantly increased investment in the resources to properly support models that won't be able to be released to the public without it.”

    Listen at 3:49

  3. ByteDance’s training run could be China’s first genuinely frontier-scale pretraining run.

    “Still, this could be the first Chinese pre-training run that's truly on the frontier.”

    Listen at 4:37

  4. ByteDance’s planned training run could be China’s first genuinely frontier-scale run.

    “This could be the first Chinese pre-training run that's truly on the frontier.”

    Listen at 4:38

  5. The training run could return ByteDance to the leading Chinese AI labs.

    “this could put ByteDance back in the conversation for leading Chinese labs.”

    Listen at 4:44

  6. Nathaniel Whittemoreon AI chip export controlsNegative6:25

    Current export controls do not prevent advanced chips being installed abroad and remotely leased to Chinese firms.

    “The export control regime only prohibits the import of advanced chips and does nothing to stop them from being installed in another country and then leased to Chinese firms.”

    Listen at 6:25

  7. Nathaniel Whittemoreon Open-source AI licensingNeutral9:04

    Open-source AI has entered an era of licensing and revenue-sharing arrangements.

    “Open source AI just entered its licensing era.”

    Listen at 9:04

  8. Nathaniel Whittemoreon Graph engineeringPositive14:28

    Graph engineering is useful for understanding the progression from prompts to contexts, harnesses, and loops.

    “the concept at least is useful to situate relative to the lineage of engineering we've had from prompt to context to harness to loop.”

    Listen at 14:28

  9. Nathaniel Whittemoreon Context engineeringPositive16:58

    Context engineering ensures AI systems can access relevant information appropriately.

    “Context engineering was all about making sure that all of that type of information was accessible in the right way.”

    Listen at 16:58

  10. Nathaniel Whittemoreon AI agents and harnessesNeutral18:29

    An AI agent consists of a model combined with its surrounding harness.

    “the agent is actually a combination of the model and the harness that surrounds it.”

    Listen at 18:29

  11. Nathaniel Whittemoreon Agent loopsPositive19:52

    Agent loops enable repeated observation, planning, action, verification, and stopping at measurable conditions.

    “Loops are the systems by which an agent can observe, plan, act, check results, and repeat until some measurable stop condition is reached.”

    Listen at 19:52

  12. Nathaniel Whittemoreon Graph engineeringPositive21:09

    Graph engineering designs interactions among agents, tools, knowledge sources, and humans.

    “Graph engineering is about designing how multiple agents, tools, knowledge sources, and humans interact and connect.”

    Listen at 21:09

  13. Nathaniel Whittemoreon Agentic graphsNeutral22:01

    Graph edges define data flows and dependencies among agent nodes.

    “The edges or interactions between those nodes define data flows and dependencies.”

    Listen at 22:01

  14. Nathaniel Whittemoreon Agentic graphsPositive22:22

    Agentic graphs specify retry, fallback, or upstream-alert behavior after node failures.

    “The graph defines what happens on failure. Is the node retried? Is it routed to a fallback? Or is there an alert upstream?”

    Listen at 22:22

  15. Nathaniel Whittemoreon AI loops and graphsNeutral22:35

    Loops govern individual agents, while graphs govern entire agentic organizations.

    “In short, a loop is how an individual agent does its job, where a graph is how an entire agentic organization works.”

    Listen at 22:35

  16. Nathaniel Whittemoreon Single-agent loopsPositive23:10

    A single loop suits sequential jobs with clear endpoints and coherent context.

    “When a single job has a clear finish line with genuinely sequential steps, and one agent's context window able to hold the whole domain, that's a good candidate for a single loop.”

    Listen at 23:10

  17. Nathaniel Whittemoreon Graph engineeringPositive23:20

    Graph engineering suits specialized, parallel, explicitly routed, resilient multi-agent workflows.

    “when the work instead splits into specialties with different handoffs, when parallelism becomes valuable, when different steps in a process want different models or tool sets, when routing has to be explicit, and when you want to design a resilient system where the failure of one node doesn't take down the rest, that's where you get into this actual graph engineering.”

    Listen at 23:20

  18. Nathaniel Whittemoreon Graph engineeringPositive23:31

    Graph engineering suits resilient systems where one node’s failure does not disable the rest.

    “when you want to design a resilient system where the failure of one node doesn't take down the rest, that's where you get into this actual graph engineering.”

    Listen at 23:31

  19. Nathaniel Whittemoreon Org graphsNeutral23:49

    Org graphs are stable, relatively permanent agentic systems.

    “Org graphs are going to be more stable agentic systems that defines a more permanent style of setup.”

    Listen at 23:49

  20. Nathaniel Whittemoreon Work graphsPositive24:41

    Work graphs can remove unnecessary tasks and spawn new ones as complexity changes.

    “An adaptive structure where tasks can disappear when evidence makes them unnecessary, or new tasks can be spawned if new complexities are discovered.”

    Listen at 24:41

  21. Nathaniel Whittemoreon Graph engineeringPositive25:23

    Graph engineering will help people reason about multi-agent systems.

    “what I think graph engineering will unlock for many is the ability to start thinking in multi-agent systems terms”

    Listen at 25:23

  22. Designing agentic systems will become an increasingly common work activity.

    “designing agentic systems is, I believe, a new work primitive, and something which we will increasingly be called upon to do.”

    Listen at 25:56

Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.

Weekly email

State of the Pod, every Monday

The week's most-discussed names and stories from podcasts, with the moments behind them. Free, five minutes, unsubscribe in one click.

Listen to the full episode and explore every guest, topic, and moment on PodLume.