diff --git a/Cargo.lock b/Cargo.lock
index ce288a7..e680cbf 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -442,6 +442,7 @@ dependencies = [
  "criterion",
  "dagrs",
  "dagx-macros",
+ "dashmap",
  "futures",
  "futures-executor",
  "parking_lot",
@@ -461,6 +462,20 @@ dependencies = [
  "syn 2.0.106",
 ]
 
+[[package]]
+name = "dashmap"
+version = "6.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5041cc499144891f3790297212f32a74fb938e5136a14943f338ef9e0ae276cf"
+dependencies = [
+ "cfg-if",
+ "crossbeam-utils",
+ "hashbrown",
+ "lock_api",
+ "once_cell",
+ "parking_lot_core",
+]
+
 [[package]]
 name = "either"
 version = "1.15.0"
@@ -641,6 +656,12 @@ dependencies = [
  "crunchy",
 ]
 
+[[package]]
+name = "hashbrown"
+version = "0.14.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e5274423e17b7c9fc20b6e7e208532f9b19825d82dfd615708b70edd83df41f1"
+
 [[package]]
 name = "hermit-abi"
 version = "0.5.2"
diff --git a/Cargo.toml b/Cargo.toml
index 55c9914..ed84368 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -22,16 +22,25 @@ default = []
 tracing = ["dep:tracing"]
 
 [lints.rust]
-unexpected_cfgs = { level = "warn", check-cfg = ['cfg(tarpaulin)', 'cfg(tarpaulin_include)'] }
+unexpected_cfgs = { level = "warn", check-cfg = [
+    'cfg(tarpaulin)',
+    'cfg(tarpaulin_include)',
+] }
 
 [dependencies]
 futures = "0.3"
 parking_lot = "0.12"
 dagx-macros = { version = "0.3.1", path = "./dagx-macros" }
 tracing = { version = "0.1", optional = true }
+dashmap = "6.1.0"
 
 [dev-dependencies]
-tokio = { version = "1", features = ["macros", "rt-multi-thread", "time", "sync"] }
+tokio = { version = "1", features = [
+    "macros",
+    "rt-multi-thread",
+    "time",
+    "sync",
+] }
 async-std = { version = "1", features = ["attributes"] }
 smol = "2"
 criterion = "0.7"
diff --git a/README.md b/README.md
index 4cc971f..2272c57 100644
--- a/README.md
+++ b/README.md
@@ -38,7 +38,7 @@ See [how it works](docs/CYCLE_PREVENTION.md).
 
 ```rust
 let sum = dag.add_task(Add).depends_on((&x, &y));
-dag.run(|fut| tokio::spawn(fut)).await?;
+dag.run().await?;
 ```
 
 That's it. No trait boilerplate, no manual channels, no node IDs.
@@ -90,7 +90,7 @@ async fn main() {
     let sum = dag.add_task(Add).depends_on((&x, &y));
 
     // Execute with true parallelism
-    dag.run(|fut| { tokio::spawn(fut); }).await.unwrap();
+    dag.run().await.unwrap();
 
     // Retrieve results
     assert_eq!(dag.get(sum).unwrap(), 5);
@@ -223,13 +223,13 @@ dagx works with any async runtime. Provide a spawner function to `run()`:
 
 ```rust
 // With Tokio
-dag.run(|fut| { tokio::spawn(fut); }).await.unwrap();
+dag.run().await.unwrap();
 
 // With async-std
-dag.run(|fut| { async_std::task::spawn(fut); }).await.unwrap();
+dag.run().await.unwrap();
 
 // With smol
-dag.run(|fut| { smol::spawn(fut).detach(); }).await.unwrap();
+dag.run().await.unwrap();
 ```
 
 ## Tutorials & Examples
diff --git a/dagx-macros/src/lib.rs b/dagx-macros/src/lib.rs
index f7d7fb5..4339bb3 100644
--- a/dagx-macros/src/lib.rs
+++ b/dagx-macros/src/lib.rs
@@ -335,7 +335,7 @@ pub fn task(_attr: TokenStream, item: TokenStream) -> TokenStream {
             quote! {
                 fn extract_and_run(
                     self,
-                    _receivers: Vec<Box<dyn std::any::Any + Send>>,
+                    _dependencies: Vec<std::sync::Arc<dyn std::any::Any + Send + Sync>>,
                 ) -> impl std::future::Future<Output = Result<Self::Output, String>> + Send {
                     async move {
                         let input = ();
@@ -350,24 +350,20 @@ pub fn task(_attr: TokenStream, item: TokenStream) -> TokenStream {
             quote! {
                 fn extract_and_run(
                     self,
-                    mut receivers: Vec<Box<dyn std::any::Any + Send>>,
+                    mut dependencies: Vec<std::sync::Arc<dyn std::any::Any + Send + Sync>>,
                 ) -> impl std::future::Future<Output = Result<Self::Output, String>> + Send {
                     async move {
-                        use futures::channel::oneshot;
                         use std::sync::Arc;
 
-                        if receivers.len() != 1 {
-                            return Err(format!("Expected 1 dependency, got {}", receivers.len()));
+                        if dependencies.len() != 1 {
+                            return Err(format!("Expected 1 dependency, got {}", dependencies.len()));
                         }
 
-                        let rx = *receivers.pop()
+                        let arc_value = dependencies.pop()
                             .unwrap()
-                            .downcast::<oneshot::Receiver<Arc<#param_type>>>()
+                            .downcast::<#param_type>()
                             .map_err(|_| format!("Type mismatch: expected Arc<{}>", std::any::type_name::<#param_type>()))?;
 
-                        let arc_value = rx.await
-                            .map_err(|_| "Channel closed before receiving value".to_string())?;
-
                         let input = (*arc_value).clone();
                         Ok(self.run(input).await)
                     }
@@ -380,51 +376,36 @@ pub fn task(_attr: TokenStream, item: TokenStream) -> TokenStream {
             let param_types: Vec<_> = params.iter().map(|(_, ty)| ty).collect();
             let indices: Vec<_> = (0..param_count).collect();
 
-            // Generate unique receiver variable names
-            let rx_vars: Vec<_> = (0..param_count)
-                .map(|i| syn::Ident::new(&format!("rx_{}", i), proc_macro2::Span::call_site()))
-                .collect();
-
             // Create syn::Index for tuple field access (avoids the suffix warning)
             let syn_indices: Vec<_> = (0..param_count).map(syn::Index::from).collect();
 
             quote! {
                 fn extract_and_run(
                     self,
-                    receivers: Vec<Box<dyn std::any::Any + Send>>,
+                    dependencies: Vec<std::sync::Arc<dyn std::any::Any + Send + Sync>>,
                 ) -> impl std::future::Future<Output = Result<Self::Output, String>> + Send {
                     async move {
-                        use futures::channel::oneshot;
                         use std::sync::Arc;
 
                         let expected_count = #param_count;
-                        if receivers.len() != expected_count {
-                            return Err(format!("Expected {} dependencies, got {}", expected_count, receivers.len()));
+                        if dependencies.len() != expected_count {
+                            return Err(format!("Expected {} dependencies, got {}", expected_count, dependencies.len()));
                         }
 
-                        let mut iter = receivers.into_iter();
+                        let mut iter = dependencies.into_iter();
 
                         // Extract each receiver
-                        #(
-                            let #rx_vars = *iter.next()
+                        let arc_results = (#(
+                            iter.next()
                                 .ok_or_else(|| format!("Missing receiver at index {}", #indices))?
-                                .downcast::<oneshot::Receiver<Arc<#param_types>>>()
+                                .downcast::<#param_types>()
                                 .map_err(|_| format!("Type mismatch at index {}: expected Arc<{}>",
-                                    #indices, std::any::type_name::<#param_types>()))?;
-                        )*
-
-                        // Await all channels concurrently
-                        let arc_results = futures::join!(
-                            #(
-                                async move {
-                                    #rx_vars.await.map_err(|_| format!("Channel {} closed", #indices))
-                                }
-                            ),*
-                        );
+                                    #indices, std::any::type_name::<#param_types>()))?,
+                        )*);
 
                         // Clone inner values and build tuple
                         let input = (#(
-                            (*arc_results.#syn_indices?).clone()
+                            (*arc_results.#syn_indices).clone()
                         ),*);
 
                         Ok(self.run(input).await)
diff --git a/docs/COMPARISONS.md b/docs/COMPARISONS.md
index ad30468..2c53ccc 100644
--- a/docs/COMPARISONS.md
+++ b/docs/COMPARISONS.md
@@ -19,7 +19,7 @@ let dag = DagRunner::new();
 let x = dag.add_task(Value(2));
 let y = dag.add_task(Value(3));
 let sum = dag.add_task(Add).depends_on((&x, &y));
-dag.run(|fut| tokio::spawn(fut)).await?;
+dag.run().await?;
 ```
 
 **dagrs** (Complex: traits + channels + IDs + manual wiring):
diff --git a/docs/CYCLE_PREVENTION.md b/docs/CYCLE_PREVENTION.md
index fbba17e..839609c 100644
--- a/docs/CYCLE_PREVENTION.md
+++ b/docs/CYCLE_PREVENTION.md
@@ -13,7 +13,7 @@ let a = dag.add_task(TaskA);
 let b = dag.add_task(TaskB);
 a.depends_on(b);
 b.depends_on(a);  // Compiles fine...
-dag.run().await?;  // ❌ Error: "Cycle detected!" (at runtime)
+dag.run()
 ```
 
 **dagx prevents this at compile time:**
diff --git a/src/builder.rs b/src/builder.rs
index d5bba4e..b4e6b40 100644
--- a/src/builder.rs
+++ b/src/builder.rs
@@ -65,7 +65,7 @@ impl IsUnitType for () {}
 /// let b = b.depends_on(&a);
 /// // Now b is a TaskHandle<i32>
 ///
-/// dag.run(|fut| { tokio::spawn(fut); }).await.unwrap();
+/// dag.run().await.unwrap();
 /// assert_eq!(dag.get(&b).unwrap(), 20);
 /// # };
 /// ```
@@ -125,7 +125,7 @@ impl<'a, Tk: Task, Deps> TaskBuilder<'a, Tk, Deps> {
     /// // Multiple dependencies: tuple form
     /// let sum = dag.add_task(Add).depends_on((&x, &y));
     ///
-    /// dag.run(|fut| { tokio::spawn(fut); }).await.unwrap();
+    /// dag.run().await.unwrap();
     /// # };
     /// ```
     pub fn depends_on<D>(self, deps: D) -> TaskHandle<Tk::Output>
diff --git a/src/error.rs b/src/error.rs
index b7afd57..7cbb01a 100644
--- a/src/error.rs
+++ b/src/error.rs
@@ -63,7 +63,7 @@ impl std::fmt::Display for DagError {
                     f,
                     "Result not found for task #{}.\n\
                      \n\
-                     Call dag.run() before accessing results with get().",
+                     Call dag.run().",
                     task_id
                 )
             }
diff --git a/src/extract.rs b/src/extract.rs
index f1b1b54..7c9f06f 100644
--- a/src/extract.rs
+++ b/src/extract.rs
@@ -11,7 +11,7 @@
 //!
 //! # How it works
 //!
-//! When a task executes, it needs to extract its inputs from type-erased oneshot receivers.
+//! When a task executes, it needs to extract its inputs from type-erased oneshot dependencies.
 //! Outputs are wrapped in Arc for efficient fanout (cheap Arc clones instead of data clones).
 //!
 //! **Modern approach (via `#[task]` macro)**:
@@ -35,8 +35,6 @@
 use std::any::Any;
 use std::sync::Arc;
 
-use futures::channel::oneshot;
-
 /// Helper trait for async input extraction from oneshot channels.
 ///
 /// Type erasure occurs only at the ExecutableNode trait boundary - by the time
@@ -47,14 +45,14 @@ use futures::channel::oneshot;
 /// which allows ANY type to work without requiring ExtractInput implementations.
 #[allow(dead_code)]
 pub(crate) trait ExtractInput: Sized + Clone {
-    fn extract_from_channels(
-        receivers: Vec<Box<dyn Any + Send>>,
+    fn extract_from_deps(
+        deps: Vec<Arc<dyn Any + Send + Sync>>,
     ) -> impl std::future::Future<Output = Result<Self, String>> + Send;
 }
 
 // Unit type - no dependencies
 impl ExtractInput for () {
-    async fn extract_from_channels(_receivers: Vec<Box<dyn Any + Send>>) -> Result<Self, String> {
+    async fn extract_from_deps(_deps: Vec<Arc<dyn Any + Send + Sync>>) -> Result<Self, String> {
         Ok(())
     }
 }
@@ -67,15 +65,14 @@ macro_rules! impl_extract_single {
     ($($t:ty),+) => {
         $(
             impl ExtractInput for $t {
-                async fn extract_from_channels(mut receivers: Vec<Box<dyn Any + Send>>) -> Result<Self, String> {
-                    if receivers.len() != 1 {
-                        return Err(format!("Expected 1 dependency, got {}", receivers.len()));
+                async fn extract_from_deps(mut dependencies: Vec<Arc<dyn Any + Send + Sync>>) -> Result<Self, String> {
+                    if dependencies.len() != 1 {
+                        return Err(format!("Expected 1 dependency, got {}", dependencies.len()));
                     }
-                    let rx = *receivers.pop()
+                    let arc_value = dependencies.pop()
                         .unwrap()
-                        .downcast::<oneshot::Receiver<Arc<$t>>>()
+                        .downcast::<$t>()
                         .map_err(|_| format!("Type mismatch: expected Arc<{}>", std::any::type_name::<$t>()))?;
-                    let arc_value = rx.await.map_err(|_| "Channel closed before receiving value".to_string())?;
                     Ok((*arc_value).clone())
                 }
             }
@@ -93,16 +90,16 @@ where
     T: 'static + Clone + Send + Sync,
     E: 'static + Clone + Send + Sync,
 {
-    async fn extract_from_channels(
-        mut receivers: Vec<Box<dyn Any + Send>>,
+    async fn extract_from_deps(
+        mut dependencies: Vec<Arc<dyn Any + Send + Sync>>,
     ) -> Result<Self, String> {
-        if receivers.len() != 1 {
-            return Err(format!("Expected 1 dependency, got {}", receivers.len()));
+        if dependencies.len() != 1 {
+            return Err(format!("Expected 1 dependency, got {}", dependencies.len()));
         }
-        let rx = *receivers
+        let arc_value = dependencies
             .pop()
             .unwrap()
-            .downcast::<oneshot::Receiver<Arc<Result<T, E>>>>()
+            .downcast::<Result<T, E>>()
             .map_err(|_| {
                 format!(
                     "Type mismatch: expected Arc<Result<{}, {}>>",
@@ -110,9 +107,6 @@ where
                     std::any::type_name::<E>()
                 )
             })?;
-        let arc_value = rx
-            .await
-            .map_err(|_| "Channel closed before receiving value".to_string())?;
         Ok((*arc_value).clone())
     }
 }
@@ -122,25 +116,22 @@ impl<T> ExtractInput for Option<T>
 where
     T: 'static + Clone + Send + Sync,
 {
-    async fn extract_from_channels(
-        mut receivers: Vec<Box<dyn Any + Send>>,
+    async fn extract_from_deps(
+        mut dependencies: Vec<Arc<dyn Any + Send + Sync>>,
     ) -> Result<Self, String> {
-        if receivers.len() != 1 {
-            return Err(format!("Expected 1 dependency, got {}", receivers.len()));
+        if dependencies.len() != 1 {
+            return Err(format!("Expected 1 dependency, got {}", dependencies.len()));
         }
-        let rx = *receivers
+        let arc_value = dependencies
             .pop()
             .unwrap()
-            .downcast::<oneshot::Receiver<Arc<Option<T>>>>()
+            .downcast::<Option<T>>()
             .map_err(|_| {
                 format!(
                     "Type mismatch: expected Arc<Option<{}>>",
                     std::any::type_name::<T>()
                 )
             })?;
-        let arc_value = rx
-            .await
-            .map_err(|_| "Channel closed before receiving value".to_string())?;
         Ok((*arc_value).clone())
     }
 }
@@ -150,25 +141,22 @@ impl<T> ExtractInput for Vec<T>
 where
     T: 'static + Clone + Send + Sync,
 {
-    async fn extract_from_channels(
-        mut receivers: Vec<Box<dyn Any + Send>>,
+    async fn extract_from_deps(
+        mut dependencies: Vec<Arc<dyn Any + Send + Sync>>,
     ) -> Result<Self, String> {
-        if receivers.len() != 1 {
-            return Err(format!("Expected 1 dependency, got {}", receivers.len()));
+        if dependencies.len() != 1 {
+            return Err(format!("Expected 1 dependency, got {}", dependencies.len()));
         }
-        let rx = *receivers
+        let arc_value = dependencies
             .pop()
             .unwrap()
-            .downcast::<oneshot::Receiver<Arc<Vec<T>>>>()
+            .downcast::<Vec<T>>()
             .map_err(|_| {
                 format!(
                     "Type mismatch: expected Arc<Vec<{}>>",
                     std::any::type_name::<T>()
                 )
             })?;
-        let arc_value = rx
-            .await
-            .map_err(|_| "Channel closed before receiving value".to_string())?;
         Ok((*arc_value).clone())
     }
 }
@@ -179,16 +167,16 @@ where
     K: 'static + Clone + Send + Sync + Eq + std::hash::Hash,
     V: 'static + Clone + Send + Sync,
 {
-    async fn extract_from_channels(
-        mut receivers: Vec<Box<dyn Any + Send>>,
+    async fn extract_from_deps(
+        mut dependencies: Vec<Arc<dyn Any + Send + Sync>>,
     ) -> Result<Self, String> {
-        if receivers.len() != 1 {
-            return Err(format!("Expected 1 dependency, got {}", receivers.len()));
+        if dependencies.len() != 1 {
+            return Err(format!("Expected 1 dependency, got {}", dependencies.len()));
         }
-        let rx = *receivers
+        let arc_value = dependencies
             .pop()
             .unwrap()
-            .downcast::<oneshot::Receiver<Arc<std::collections::HashMap<K, V>>>>()
+            .downcast::<std::collections::HashMap<K, V>>()
             .map_err(|_| {
                 format!(
                     "Type mismatch: expected Arc<HashMap<{}, {}>>",
@@ -196,9 +184,6 @@ where
                     std::any::type_name::<V>()
                 )
             })?;
-        let arc_value = rx
-            .await
-            .map_err(|_| "Channel closed before receiving value".to_string())?;
         Ok((*arc_value).clone())
     }
 }
@@ -209,25 +194,22 @@ impl<T> ExtractInput for Arc<T>
 where
     T: 'static + Clone + Send + Sync,
 {
-    async fn extract_from_channels(
-        mut receivers: Vec<Box<dyn Any + Send>>,
+    async fn extract_from_deps(
+        mut dependencies: Vec<Arc<dyn Any + Send + Sync>>,
     ) -> Result<Self, String> {
-        if receivers.len() != 1 {
-            return Err(format!("Expected 1 dependency, got {}", receivers.len()));
+        if dependencies.len() != 1 {
+            return Err(format!("Expected 1 dependency, got {}", dependencies.len()));
         }
-        let rx = *receivers
+        let arc_arc_value = dependencies
             .pop()
             .unwrap()
-            .downcast::<oneshot::Receiver<Arc<Arc<T>>>>()
+            .downcast::<Arc<T>>()
             .map_err(|_| {
                 format!(
                     "Type mismatch: expected Arc<Arc<{}>>",
                     std::any::type_name::<T>()
                 )
             })?;
-        let arc_arc_value = rx
-            .await
-            .map_err(|_| "Channel closed before receiving value".to_string())?;
         // Return the inner Arc (clone is cheap - just refcount increment)
         Ok(Arc::clone(&*arc_arc_value))
     }
@@ -251,43 +233,32 @@ where
 macro_rules! impl_extract_tuple {
     ($($T:ident : $idx:tt),+) => {
         impl<$($T: 'static + Clone + Send + Sync),+> ExtractInput for ($($T,)+) {
-            async fn extract_from_channels(receivers: Vec<Box<dyn Any + Send>>) -> Result<Self, String> {
-                // Verify we have the correct number of receivers
+            async fn extract_from_deps(dependencies: Vec<Arc<dyn Any + Send + Sync>>) -> Result<Self, String> {
+                // Verify we have the correct number of dependencies
                 let expected_count = [$($idx),+].len();
-                if receivers.len() != expected_count {
-                    return Err(format!("Expected {} dependencies, got {}", expected_count, receivers.len()));
+                if dependencies.len() != expected_count {
+                    return Err(format!("Expected {} dependencies, got {}", expected_count, dependencies.len()));
                 }
 
                 // Convert Vec to iterator so we can consume it
-                let mut iter = receivers.into_iter();
+                let mut iter = dependencies.into_iter();
 
-                // Extract each receiver by consuming from the iterator
-                // Receivers now contain Arc-wrapped values for efficient fanout
+                // dependencies now contain Arc-wrapped values for efficient fanout
                 // Dereference Box after downcasting to get the receiver itself
                 // Use type names as variable names to avoid numeric suffix issues
-                $(
-                    #[allow(non_snake_case)]
-                    let $T = *iter.next()
-                        .ok_or_else(|| format!("Missing receiver at index {}", $idx))?
-                        .downcast::<oneshot::Receiver<Arc<$T>>>()
-                        .map_err(|_| format!("Type mismatch at index {}: expected Arc<{}>",
-                            $idx, std::any::type_name::<$T>()))?;
-                )+
-
-                // Await all channels concurrently - this is a key optimization!
-                // We can start receiving as soon as ANY dependency completes, not waiting
-                // for them sequentially.
-                let arc_results = futures::join!(
+                let arc_results = (
                     $(
-                        async move {
-                            $T.await.map_err(|_| format!("Channel {} closed", $idx))
-                        }
-                    ),+
+                            iter.next()
+                            .ok_or_else(|| format!("Missing receiver at index {}", $idx))?
+                            .downcast::<$T>()
+                            .map_err(|_| format!("Type mismatch at index {}: expected Arc<{}>",
+                                $idx, std::any::type_name::<$T>()))?,
+                    )+
                 );
 
                 // Clone inner values from Arc (cheap for small types, necessary for ownership)
                 Ok(($(
-                    (*arc_results.$idx?).clone()
+                    (*arc_results.$idx).clone()
                 ,)+))
             }
         }
diff --git a/src/lib.rs b/src/lib.rs
index 5a4660d..903fe3b 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -52,7 +52,7 @@
 //! let sum = dag.add_task(Add).depends_on((&x, &y));
 //!
 //! // Execute and retrieve results
-//! dag.run(|fut| { tokio::spawn(fut); }).await.unwrap();
+//! dag.run().await.unwrap();
 //! assert_eq!(dag.get(sum).unwrap(), 5);
 //! # };
 //! ```
@@ -322,7 +322,7 @@
 //! let plus1 = dag.add_task(Add(1)).depends_on(&base);
 //! let times2 = dag.add_task(Scale(2)).depends_on(&base);
 //!
-//! dag.run(|fut| { tokio::spawn(fut); }).await.unwrap();
+//! dag.run().await.unwrap();
 //!
 //! assert_eq!(dag.get(plus1).unwrap(), 11);
 //! assert_eq!(dag.get(times2).unwrap(), 20);
@@ -378,7 +378,7 @@
 //! let active = dag.add_task(Active(true));
 //! let result = dag.add_task(FormatUser).depends_on((&name, &age, &active));
 //!
-//! dag.run(|fut| { tokio::spawn(fut); }).await.unwrap();
+//! dag.run().await.unwrap();
 //!
 //! assert_eq!(dag.get(result).unwrap(), "User: Alice, Age: 30, Active: true");
 //! # };
@@ -430,7 +430,7 @@
 //! // Layer 3: Final result
 //! let total = dag.add_task(Add).depends_on((&sum_xy, &prod_yz)); // 5 + 15 = 20
 //!
-//! dag.run(|fut| { tokio::spawn(fut); }).await.unwrap();
+//! dag.run().await.unwrap();
 //!
 //! assert_eq!(dag.get(total).unwrap(), 20);
 //! # };
@@ -473,7 +473,7 @@
 //! let user = dag.add_task(CreateUser);
 //! let formatted = dag.add_task(FormatUser).depends_on(&user);
 //!
-//! dag.run(|fut| { tokio::spawn(fut); }).await.unwrap();
+//! dag.run().await.unwrap();
 //!
 //! assert_eq!(dag.get(formatted).unwrap(), "Alice is 30 years old");
 //! # };
@@ -539,11 +539,11 @@
 //! # let dag = DagRunner::new();
 //! # let node = dag.add_task(Value(42));
 //! // Simple approach with .unwrap()
-//! dag.run(|fut| { tokio::spawn(fut); }).await.unwrap();
+//! dag.run().await.unwrap();
 //! let result = dag.get(node).unwrap();
 //!
 //! // Or handle errors explicitly
-//! match dag.run(|fut| { tokio::spawn(fut); }).await {
+//! match dag.run().await {
 //!     Ok(_) => println!("DAG executed successfully"),
 //!     Err(e) => eprintln!("DAG execution failed: {}", e),
 //! }
@@ -656,7 +656,7 @@
 //! let task2 = dag.add_task(ProcessData).depends_on(&data);
 //! let task3 = dag.add_task(ProcessData).depends_on(&data);
 //!
-//! dag.run(|fut| { tokio::spawn(fut); }).await.unwrap();
+//! dag.run().await.unwrap();
 //! # };
 //! ```
 //!
diff --git a/src/node.rs b/src/node.rs
index ee0a24c..41d3b27 100644
--- a/src/node.rs
+++ b/src/node.rs
@@ -11,18 +11,10 @@ use std::future::Future;
 use std::pin::Pin;
 use std::sync::Arc;
 
-use futures::channel::oneshot;
-
 use crate::error::DagError;
 use crate::task::Task;
 use crate::types::NodeId;
 
-/// Type-erased channel endpoints (senders or receivers)
-type ErasedChannels = Vec<Box<dyn Any + Send>>;
-
-/// Type-erased channel pair (senders, receivers)
-type ChannelPair = (ErasedChannels, ErasedChannels);
-
 /// Type-erased async execution result (Arc-wrapped for efficient fanout)
 type ExecuteFuture =
     Pin<Box<dyn Future<Output = Result<Arc<dyn Any + Send + Sync>, DagError>> + Send>>;
@@ -38,21 +30,13 @@ type ExecuteFuture =
 ///
 /// This is an implementation detail and not part of the public API.
 pub(crate) trait ExecutableNode: Send {
-    /// Create type-erased output channels for this node's dependents.
-    ///
-    /// Returns (senders, receivers) where:
-    /// - senders: Type-erased senders that this node will use to send outputs
-    /// - receivers: Type-erased receivers that dependents will use to receive inputs
-    fn create_output_channels(&self, num_dependents: usize) -> ChannelPair;
-
     /// Execute the task with type-erased input receivers and output senders.
     ///
     /// The TypedNode implementation knows the concrete types and can safely downcast.
     /// Consumes the node and returns the output value.
-    fn execute_with_channels(
+    fn execute_with_deps(
         self: Box<Self>,
-        receivers: ErasedChannels,
-        senders: ErasedChannels,
+        dependencies: Vec<Arc<dyn Any + Send + Sync>>,
     ) -> ExecuteFuture;
 }
 
@@ -86,30 +70,15 @@ where
     T::Input: 'static + Clone,
     T::Output: 'static + Clone,
 {
-    fn create_output_channels(&self, num_dependents: usize) -> ChannelPair {
-        let mut senders = Vec::new();
-        let mut receivers = Vec::new();
-
-        for _ in 0..num_dependents {
-            // Create channels that pass Arc-wrapped outputs (efficient for fanout)
-            let (tx, rx) = oneshot::channel::<Arc<T::Output>>();
-            senders.push(Box::new(tx) as Box<dyn Any + Send>);
-            receivers.push(Box::new(rx) as Box<dyn Any + Send>);
-        }
-
-        (senders, receivers)
-    }
-
-    fn execute_with_channels(
+    fn execute_with_deps(
         self: Box<Self>,
-        receivers: ErasedChannels,
-        senders: ErasedChannels,
+        dependencies: Vec<Arc<dyn Any + Send + Sync>>,
     ) -> ExecuteFuture {
         Box::pin(async move {
             // Use the task's extract_and_run method which has inline extraction logic
             // generated by the #[task] macro. This allows ANY type to work without
             // requiring ExtractInput trait implementations.
-            let output = match self.task.extract_and_run(receivers).await {
+            let output = match self.task.extract_and_run(dependencies).await {
                 Ok(output) => output,
                 Err(_msg) => {
                     // Failed to extract input - likely channel closed or type mismatch
@@ -124,17 +93,6 @@ where
             // Downstream tasks receive &T after extraction from Arc.
             let arc_output = Arc::new(output);
 
-            // Send Arc to all dependents via their channels
-            // TypedNode<T> knows T::Output, so downcast is safe
-            for sender_any in senders {
-                if let Ok(sender) = sender_any.downcast::<oneshot::Sender<Arc<T::Output>>>() {
-                    // Clone Arc (just atomic refcount increment, NOT data clone)
-                    // This is the key performance optimization for fan-out patterns
-                    // Ignore send errors - dependent might have been cancelled
-                    let _ = sender.send(Arc::clone(&arc_output));
-                }
-            }
-
             // Return Arc-wrapped output (for storage by runner)
             Ok(arc_output as Arc<dyn Any + Send + Sync>)
         })
diff --git a/src/runner.rs b/src/runner.rs
index 65258bb..c1ab82c 100644
--- a/src/runner.rs
+++ b/src/runner.rs
@@ -5,6 +5,7 @@
 //!
 //! Uses Mutex for interior mutability to enable builder pattern (`&self` instead of `&mut self`).
 
+use std::any::Any;
 use std::collections::{HashMap, HashSet, VecDeque};
 use std::panic::AssertUnwindSafe;
 use std::sync::{
@@ -12,9 +13,9 @@ use std::sync::{
     Arc,
 };
 
-use futures::channel::mpsc;
-use futures::future::BoxFuture;
-use futures::{FutureExt, StreamExt};
+use dashmap::DashMap;
+use futures::stream::FuturesUnordered;
+use futures::{FutureExt, StreamExt, TryFutureExt};
 use parking_lot::Mutex;
 
 #[cfg(feature = "tracing")]
@@ -82,7 +83,7 @@ impl<'a> Drop for RunGuard<'a> {
 /// let y = dag.add_task(LoadValue::new(3));
 /// let sum = dag.add_task(Add).depends_on((&x, &y));
 ///
-/// dag.run(|fut| { tokio::spawn(fut); }).await.unwrap();
+/// dag.run().await.unwrap();
 ///
 /// assert_eq!(dag.get(sum).unwrap(), 5);
 /// # };
@@ -95,7 +96,7 @@ impl<'a> Drop for RunGuard<'a> {
 /// Outputs are Arc-wrapped and stored separately for retrieval via get().
 /// Arc enables efficient sharing during fanout without cloning data.
 pub struct DagRunner {
-    pub(crate) nodes: Mutex<Vec<Option<Box<dyn ExecutableNode>>>>,
+    pub(crate) nodes: Mutex<Vec<Option<Box<dyn ExecutableNode + Sync>>>>,
     pub(crate) outputs: Mutex<HashMap<NodeId, std::sync::Arc<dyn std::any::Any + Send + Sync>>>,
     pub(crate) edges: Mutex<HashMap<NodeId, Vec<NodeId>>>, // node -> dependencies
     pub(crate) dependents: Mutex<HashMap<NodeId, Vec<NodeId>>>, // node -> tasks that depend on it
@@ -188,13 +189,13 @@ impl DagRunner {
     /// // Construct task with offset of 1
     /// let inc = dag.add_task(AddOffset::new(1)).depends_on(&base);
     ///
-    /// dag.run(|fut| { tokio::spawn(fut); }).await.unwrap();
+    /// dag.run().await.unwrap();
     /// assert_eq!(dag.get(&inc).unwrap(), 11);
     /// # };
     /// ```
     pub fn add_task<Tk>(&self, task: Tk) -> TaskBuilder<'_, Tk, Pending>
     where
-        Tk: Task + 'static,
+        Tk: Task + Sync + 'static,
         Tk::Input: 'static + Clone,
         Tk::Output: 'static + Clone,
     {
@@ -272,7 +273,7 @@ impl DagRunner {
     /// let b = dag.add_task(Value(2));
     /// let sum = dag.add_task(Add).depends_on((&a, &b));
     ///
-    /// dag.run(|fut| { tokio::spawn(fut); }).await.unwrap(); // Executes all tasks
+    /// dag.run().await.unwrap(); // Executes all tasks
     /// # };
     /// ```
     ///
@@ -283,11 +284,8 @@ impl DagRunner {
     /// Type erasure occurs only at the ExecutableNode trait boundary - channels are created
     /// with full type information and type-erased just before passing to execute_with_channels.
     #[inline]
-    #[cfg_attr(feature = "tracing", tracing::instrument(skip(self, spawner)))]
-    pub async fn run<S>(&self, spawner: S) -> DagResult<()>
-    where
-        S: Fn(BoxFuture<'static, ()>),
-    {
+    #[cfg_attr(feature = "tracing", tracing::instrument(skip(self)))]
+    pub async fn run(&self) -> DagResult<()> {
         #[cfg(feature = "tracing")]
         info!("starting DAG execution");
         // Acquire run lock to prevent concurrent executions
@@ -320,47 +318,8 @@ impl DagRunner {
 
         let edges = self.edges.lock().clone();
 
-        // Create all channels upfront
-        // Map: (producer_id, consumer_id) -> receiver index
-        // We store receivers in a Vec per consumer to maintain order
-        let mut consumer_receivers: HashMap<NodeId, Vec<Box<dyn std::any::Any + Send>>> =
-            HashMap::new();
-        let mut producer_senders: HashMap<NodeId, Vec<Box<dyn std::any::Any + Send>>> =
-            HashMap::new();
-
-        // Create channels for each producer-consumer relationship
-        // We need to lock nodes temporarily to call create_output_channels
-        {
-            let nodes_lock = self.nodes.lock();
-
-            for (consumer_id, producer_ids) in &edges {
-                for &producer_id in producer_ids {
-                    // Get the producer node (it's still in the vector at this point)
-                    if let Some(Some(producer_node)) = nodes_lock.get(producer_id.0) {
-                        // Ask the producer to create ONE channel for this consumer
-                        let (mut senders, mut receivers) = producer_node.create_output_channels(1);
-
-                        // Store the sender for the producer
-                        producer_senders
-                            .entry(producer_id)
-                            .or_default()
-                            .push(senders.pop().unwrap());
-
-                        // Store the receiver for the consumer (order matters!)
-                        consumer_receivers
-                            .entry(*consumer_id)
-                            .or_default()
-                            .push(receivers.pop().unwrap());
-                    }
-                }
-            }
-        }
-
-        // Create a channel to collect Arc-wrapped outputs from all tasks
-        let (output_tx, mut output_rx) = mpsc::unbounded::<(
-            NodeId,
-            Result<std::sync::Arc<dyn std::any::Any + Send + Sync>, DagError>,
-        )>();
+        let outputs: DashMap<NodeId, Arc<dyn Any + Send + Sync>> = DashMap::new();
+        let mut first_error = None;
 
         for layer in layers {
             #[cfg(feature = "tracing")]
@@ -402,8 +361,6 @@ impl DagRunner {
                     "executing task inline (single-task layer optimization)"
                 );
 
-                let out_tx = output_tx.clone();
-
                 // Take ownership of the node
                 let node = {
                     let mut nodes_lock = self.nodes.lock();
@@ -411,11 +368,11 @@ impl DagRunner {
                 };
 
                 if let Some(node) = node {
-                    // Take the receivers for this consumer
-                    let receivers = consumer_receivers.remove(&node_id).unwrap_or_default();
-
-                    // Take the senders for this producer
-                    let senders = producer_senders.remove(&node_id).unwrap_or_default();
+                    let dependencies: Vec<_> = edges[&node_id]
+                        .iter()
+                        .flat_map(|dep| outputs.get(dep))
+                        .map(|value| value.clone())
+                        .collect();
 
                     // Execute inline with panic handling.
                     //
@@ -423,7 +380,7 @@ impl DagRunner {
                     // DagError::TaskPanicked, matching the behavior of async runtimes when
                     // they spawn tasks. This guarantees consistent error handling whether
                     // a task executes inline (single-task layer) or spawned (multi-task layer).
-                    let result = AssertUnwindSafe(node.execute_with_channels(receivers, senders))
+                    let result = AssertUnwindSafe(node.execute_with_deps(dependencies))
                         .catch_unwind()
                         .await
                         .unwrap_or_else(|panic_payload| {
@@ -451,79 +408,94 @@ impl DagRunner {
                         });
 
                     // Send the output (ignore send errors - receiver may be dropped)
-                    let _ = out_tx.unbounded_send((node_id, result));
+                    match result {
+                        Ok(output) => {
+                            outputs.insert(node_id, output);
+                        }
+                        Err(e) => {
+                            first_error.get_or_insert(e);
+                        }
+                    }
                 }
             } else {
                 // Slow path: Multiple tasks require spawning and coordination
-                // Create a channel to track task completion for this layer
-                let (tx, mut rx) = mpsc::channel::<()>(layer.len());
-
                 // Spawn each task in this layer
-                for &node_id in &layer {
-                    #[cfg(feature = "tracing")]
-                    trace!(task_id = node_id.0, "spawning task");
-                    let mut task_tx = tx.clone();
-                    let out_tx = output_tx.clone();
+                let outputs = &outputs;
+                let mut futures: FuturesUnordered<_> = FuturesUnordered::new();
+                layer
+                    .into_iter()
+                    .flat_map(|node_id| {
+                        #[cfg(feature = "tracing")]
+                        trace!(task_id = node_id.0, "spawning task");
 
-                    // Take ownership of the node
-                    let node = {
-                        let mut nodes_lock = self.nodes.lock();
-                        nodes_lock[node_id.0].take()
-                    };
-
-                    if let Some(node) = node {
-                        // Take the receivers for this consumer
-                        let receivers = consumer_receivers.remove(&node_id).unwrap_or_default();
-
-                        // Take the senders for this producer
-                        let senders = producer_senders.remove(&node_id).unwrap_or_default();
-
-                        // Create a 'static future by taking ownership
-                        let inner_future = async move {
-                            let task_future = node.execute_with_channels(receivers, senders);
-
-                            let result = task_future.await;
-
-                            // Send the output (ignore send errors - receiver may be dropped)
-                            let _ = out_tx.unbounded_send((node_id, result));
-
-                            // Signal completion (ignore send errors - receiver may be dropped)
-                            let _ = task_tx.try_send(());
+                        // Take ownership of the node
+                        let node = {
+                            let mut nodes_lock = self.nodes.lock();
+                            nodes_lock[node_id.0].take()
                         };
+                        if let Some(node) = node {
+                            let dependencies: Vec<_> = edges[&node_id]
+                                .iter()
+                                .flat_map(|dep| outputs.get(dep))
+                                .map(|value| value.clone())
+                                .collect();
 
-                        // Spawn the task using the provided spawner
-                        spawner(Box::pin(inner_future));
+                            // Spawn the task using the provided spawner
+                            let inner_future = async move {
+                                let result = node.execute_with_deps(dependencies).await?;
+                                outputs.insert(node_id, result);
+                                Ok(())
+                            };
+
+                            Some(
+                                AssertUnwindSafe(Box::pin(inner_future))
+                                    .catch_unwind()
+                                    .unwrap_or_else(move |panic_payload| {
+                                        // Convert panic to error
+                                        let panic_message =
+                                            if let Some(s) = panic_payload.downcast_ref::<&str>() {
+                                                s.to_string()
+                                            } else if let Some(s) =
+                                                panic_payload.downcast_ref::<String>()
+                                            {
+                                                s.clone()
+                                            } else {
+                                                "unknown panic".to_string()
+                                            };
+
+                                        #[cfg(feature = "tracing")]
+                                        error!(
+                                            task_id = node_id.0,
+                                            panic_message = %panic_message,
+                                            "task panicked during inline execution"
+                                        );
+
+                                        Err(DagError::TaskPanicked {
+                                            task_id: node_id.0,
+                                            panic_message,
+                                        })
+                                    }),
+                            )
+                        } else {
+                            None
+                        }
+                    })
+                    .for_each(|fut| futures.push(fut));
+
+                while let Some(out) = futures.next().await {
+                    if let Err(e) = out {
+                        first_error.get_or_insert(e);
                     }
                 }
-
-                // Drop the original sender so the channel closes when all tasks complete
-                drop(tx);
-
-                // Wait for all tasks in this layer to complete
-                while rx.next().await.is_some() {}
-            }
-        }
-
-        // Drop the output sender so the channel closes
-        drop(output_tx);
-
-        // Collect all outputs and check for errors
-        let mut collected = Vec::new();
-        let mut first_error = None;
-        while let Some((node_id, result)) = output_rx.next().await {
-            match result {
-                Ok(output) => collected.push((node_id, output)),
-                Err(e) if first_error.is_none() => first_error = Some(e),
-                Err(_) => {} // Ignore subsequent errors
             }
         }
 
         // Insert all successful outputs (avoiding holding lock across await)
-        let mut outputs = self.outputs.lock();
-        for (node_id, output) in collected {
-            outputs.insert(node_id, output);
+        let mut outputs_lock = self.outputs.lock();
+        for (node_id, output) in outputs {
+            outputs_lock.insert(node_id, output);
         }
-        drop(outputs);
+        drop(outputs_lock);
 
         // Return first error if any
         if let Some(err) = first_error {
@@ -575,7 +547,7 @@ impl DagRunner {
     /// // Construct task with specific setting value
     /// let task = dag.add_task(Configuration::new(42));
     ///
-    /// dag.run(|fut| { tokio::spawn(fut); }).await.unwrap();
+    /// dag.run().await.unwrap();
     ///
     /// assert_eq!(dag.get(task).unwrap(), 42);
     /// # };
diff --git a/src/task.rs b/src/task.rs
index f362e87..54ee361 100644
--- a/src/task.rs
+++ b/src/task.rs
@@ -3,7 +3,7 @@
 //! This module defines the core Task trait that all DAG nodes must implement,
 //! along with convenience functions for creating tasks from closures.
 
-use std::future::Future;
+use std::{future::Future, sync::Arc};
 
 /// A unit of async work with typed inputs and outputs.
 ///
@@ -106,7 +106,7 @@ pub trait Task: Send {
     #[doc(hidden)]
     fn extract_and_run(
         self,
-        receivers: Vec<Box<dyn std::any::Any + Send>>,
+        dependencies: Vec<Arc<dyn std::any::Any + Send + Sync>>,
     ) -> impl Future<Output = Result<Self::Output, String>> + Send;
 }
 
@@ -149,10 +149,10 @@ where
 
         async fn extract_and_run(
             self,
-            receivers: Vec<Box<dyn std::any::Any + Send>>,
+            dependencies: Vec<Arc<dyn std::any::Any + Send + Sync>>,
         ) -> Result<Self::Output, String> {
             // For task_fn, we use ExtractInput since we don't have macro-generated extraction
-            let input = I::extract_from_channels(receivers)
+            let input = I::extract_from_deps(dependencies)
                 .await
                 .map_err(|e| format!("Failed to extract input: {}", e))?;
 
diff --git a/src/types.rs b/src/types.rs
index cca4e5d..7968d60 100644
--- a/src/types.rs
+++ b/src/types.rs
@@ -33,7 +33,7 @@ pub(crate) struct NodeId(pub usize);
 /// let dag = DagRunner::new();
 /// let node = dag.add_task(LoadValue::new(42));
 ///
-/// dag.run(|fut| { tokio::spawn(fut); }).await.unwrap();
+/// dag.run().await.unwrap();
 ///
 /// assert_eq!(dag.get(node).unwrap(), 42);
 /// # };