P E E R T O P E E R M A G A Z I N E ยท S U M M E R 2 0 2 5
11
This evolution has been
primarily enabled by
advances in generative
AI and large language
models (LLMs). Early LLM
applications (e.g., ChatGPT
in 2022) were reactive
assistants that produced
content only when prompted.
Newer systems combine
LLMs with tool use, memory,
and goal-driven planning,
effectively "autonomizing"
the AI (https://www.credo.
ai/recourseslongform/
from-assistant-to-agent-
navigating-the-governance-
challenges-of-increasingly-
autonomous-ai). These
agents can dynamically call
APIs, write and execute code,
perform analysis, query data
sources, and adjust plans
in real-time. In essence,
the AI can now figure out
how to achieve an assigned
objective, much like a junior
employee would, rather
than requiring the human
to micromanage each step.
This new agentic paradigm
promises substantial
benefits; one analysis
predicts greater workforce
specialization, improved
information accuracy, and
accelerated innovation as
draft a clause, which the
lawyer then edits (AI is
helping but under continuous
supervision). Delegate mode
would mean the lawyer
instead asks an AI agent to
review an entire contract and
flag issues, or even to handle
a first round of negotiations,
with the AI deciding when to
loop the human back in.
Microsoft CEO Satya Nadella
captures this difference in
evolving developer tools.
Initially, AI was a "pair
programmer" (working
alongside, step-by-step),
but it is becoming more of
a "peer programmer" that
can independently write
and refactor code across
a codebase. He envisions
knowledge workers
supported by "fleets of AI
agents," such as researcher
agents, analyst agents, and
coder agents, some of which
work under close human
direction. In contrast, others
are fully delegated to carry
out tasks independently.
Nadella suggests users will
sometimes stay in the loop
and sometimes delegate
entirely, which will require
new ways to coordinate and
AI agents take on routine
complexities (https://hbr.
org/2024/12/what-is-agentic-
ai-and-how-will-it-change-
work). Early research even
indicates human-AI teams
can perform best when the
AI takes the lead role and
delegates tasks to humans,
rather than vice versa.
"ASSIST MODE" VS
"DELEGATE MODE"
Frontier organizations are
distinguishing between AI in
assist roles versus delegate
roles. In assist mode, AI
acts as a capable helper
or copilot, always at the
human's side, but waiting for
instructions and feedback.
Humans are the ultimate
decision-makers, and the
AI's autonomy is limited to
providing suggestions or
taking single-step actions.
In delegate mode, a human
can assign a goal or task
to an AI agent and allow it
to execute autonomously,
intervening only if needed.
In practical terms, assist
mode might mean a lawyer
using an AI assistant to
track what these agents do.
In short, assist mode keeps
the human in charge of each
action, while delegate mode
entrusts the AI agent to
drive parts of the workflow,
with the human stepping
back into a supervisory or
exception-handling role
(https://www.geekwire.
com/2025/microsoft-ceo-
satya-nadella-on-the-tech-
giants-50th-anniversary-
and-what-comes-next/).
EXAMPLES OF AI-
DRIVEN DELEGATION
IN PRACTICE
Management Consulting
& Project Management:
In consulting, AI agents
are beginning to handle
complex coordination and
research tasks. For example,
McKinsey's tool Lilli is used
by over 75% of employees
to create PowerPoints
and take over some of the
work junior employees and
analysts previously did
(https://www.entrepreneur.
com/business-news/
ai-creates-powerpoints-
at-mckinsey-replacing-
junior-workers/492624).
Similarly, at Microsoft
Ignite 2024, Nadella
showcased upcoming
"Project Manager" AI
agents that autonomously
manage project timelines,
task assignments, and
resource allocation
(LegalTech company Lupl
is working on the same
for law firms, a magnetic
Legal Project Manager). In
essence, such an agent can
monitor project progress
Early research even indicates human-AI teams
can perform best when the AI takes the lead
role and delegates tasks to humans, rather
than vice versa.
