diff --git a/src/autowiring/CoreJob.cpp b/src/autowiring/CoreJob.cpp
index 3737167af..78a07a70b 100644
--- a/src/autowiring/CoreJob.cpp
+++ b/src/autowiring/CoreJob.cpp
@@ -52,8 +52,10 @@ void CoreJob::OnPended(std::unique_lock<std::mutex>&& lk){
     // Need to ask the thread pool to handle our events again:
     m_curEventInTeardown = false;
 
-    if (m_curEvent)
-      delete static_cast<std::future<void>*>(m_curEvent);
+    std::future<void>* future = static_cast<std::future<void>*>(std::atomic_exchange<void*>(&m_curEvent, nullptr));
+    if (future) {
+      delete future;
+    }
 
     m_curEvent = new std::future<void>(
       std::async(
@@ -97,7 +99,7 @@ bool CoreJob::OnStart(void) {
 
   m_running = true;
 
-  std::unique_lock<std::mutex> lk;
+  std::unique_lock<std::mutex> lk(m_dispatchLock);
   if(m_pHead)
     // Simulate a pending event, because we need to set up our async:
     OnPended(std::move(lk));
@@ -122,21 +124,20 @@ void CoreJob::OnStop(bool graceful) {
 }
 
 void CoreJob::DoAdditionalWait(void) {
-  if (m_curEvent) {
-    std::future<void>* ptr = static_cast<std::future<void>*>(m_curEvent);
-    ptr->wait();
-    delete ptr;
-    m_curEvent = nullptr;
+  std::future<void>* future = static_cast<std::future<void>*>(std::atomic_exchange<void*>(&m_curEvent, nullptr));
+
+  if (future) {
+    future->wait();
+    delete future;
   }
 }
 
 bool CoreJob::DoAdditionalWait(std::chrono::nanoseconds timeout) {
-  if (!m_curEvent)
+  std::future<void>* future = static_cast<std::future<void>*>(std::atomic_exchange<void*>(&m_curEvent, nullptr));
+  if (!future)
     return true;
 
-  std::future<void>* ptr = static_cast<std::future<void>*>(m_curEvent);
-  auto status = ptr->wait_for(NanosecondsForFutureWait(timeout));
-  delete ptr;
-  m_curEvent = nullptr;
+  const auto status = future->wait_for(NanosecondsForFutureWait(timeout));
+  delete future;
   return status == std::future_status::ready;
 }
diff --git a/src/autowiring/CoreJob.h b/src/autowiring/CoreJob.h
index c1e0c7b50..700870971 100644
--- a/src/autowiring/CoreJob.h
+++ b/src/autowiring/CoreJob.h
@@ -18,7 +18,7 @@ class CoreJob:
   bool m_running = false;
 
   // The current outstanding async in the thread pool, if one exists:
-  void* m_curEvent = nullptr;
+  std::atomic<void*> m_curEvent{ nullptr };
 
   // Flag, indicating whether curEvent is in a teardown pathway.  This
   // flag is highly stateful.
diff --git a/src/autowiring/DispatchQueue.h b/src/autowiring/DispatchQueue.h
index 54004fb5e..cfc55098f 100644
--- a/src/autowiring/DispatchQueue.h
+++ b/src/autowiring/DispatchQueue.h
@@ -406,9 +406,10 @@ class DispatchQueue {
     // Create the thunk first to reduce the amount of time we spend in lock:
     auto thunk = new autowiring::DispatchThunk<_Fx>(std::forward<_Fx>(fx));
 
-    m_dispatchLock.lock();
+    std::unique_lock<std::mutex> lk(m_dispatchLock);
+
     if (m_count >= m_dispatchCap) {
-      m_dispatchLock.unlock();
+      lk.unlock();
       delete thunk;
       return false;
     }
@@ -420,16 +421,15 @@ class DispatchQueue {
     if (m_pHead) {
       m_pTail->m_pFlink = thunk;
       m_pTail = thunk;
-      m_dispatchLock.unlock();
+      // Notification as needed:
+      OnPended(std::move(lk));
     }
     else {
       m_pHead = m_pTail = thunk;
-      m_dispatchLock.unlock();
+      // Notification as needed:
+      OnPended(std::move(lk));
       m_queueUpdated.notify_all();
     }
-
-    // Notification as needed:
-    OnPended(std::unique_lock<std::mutex>{});
     return true;
   }
 };
diff --git a/src/autowiring/test/CoreJobTest.cpp b/src/autowiring/test/CoreJobTest.cpp
index 7cca8d9e2..0a436ad0e 100644
--- a/src/autowiring/test/CoreJobTest.cpp
+++ b/src/autowiring/test/CoreJobTest.cpp
@@ -2,6 +2,7 @@
 #include "stdafx.h"
 #include <autowiring/CoreJob.h>
 #include THREAD_HEADER
+#include ARRAY_HEADER
 
 class CoreJobTest:
   public testing::Test
@@ -179,3 +180,28 @@ TEST_F(CoreJobTest, RecursiveDeadlock) {
     std::this_thread::sleep_for(std::chrono::milliseconds(100));
   };
 }
+
+TEST_F(CoreJobTest, PendFromMultipleThreads) {
+  AutoCurrentContext ctxt;
+  AutoRequired<CoreJob> cj;
+  std::array<std::thread, 4> threads;
+  const size_t times{ 256 };
+  int counter{ 0 };
+
+  for (size_t i = 0; i < threads.size(); i++) {
+    threads[i] = std::thread([&] {
+      ctxt->DelayUntilInitiated();
+      for (int j = 0; j < times; j++) {
+        *cj += [&counter] {
+          counter++; // Should be updated exclusively in the CoreJob's thread
+        };
+      }
+    });
+  }
+  ctxt->Initiate();
+  for (size_t i = 0; i < threads.size(); i++) {
+    threads[i].join();
+  }
+  ctxt->SignalShutdown(true);
+  ASSERT_EQ(times*threads.size(), counter);
+}