Your AI strategy has a trust problem, not a tooling problem

Elena Verna (cross-posted by Lenny Rachitsky) · email · 2026-06-05 Reader 🖥️

Agentic AI AI Confidence Theater Enterprise AI Adoption & Readiness Hierarchy & Org Charts Agency Human as Orchestrator of Agents Team Autonomy Trust

AI tooling produces no velocity unless companies first dismantle the title-gated access, decision-bottlenecking, and middle-management telephone-game that signal distrust of employees — the real unlock is employee agency, not AI agents.

practical · argumentative — Elena is arguing for an action: rewire your org so employees have agency, not just AI tools. The piece moves from diagnosis (command-and-control kills velocity) to prescription (prototype a flat, high-autonomy unit) to a warning (employees must also choose to bear accountability).

Structure

Part	Role
Opening frame: 'Agency > Agents'	Thesis statement — names the binary the rest of the piece will defend
Diagnosis: how 'protecting' employees actually traps them (title-gated access, boss-only decisions)	Evidence — illustrates the structural failure being named
Counter-model: faster + cheaper decisions when context is ungated (Anthropic 'member of technical staff' as proof point)	Constructive case — what the alternative produces
Management bloat: middle-management as a structural symptom of distrust	Elaboration — names what becomes redundant when agency is distributed
Prescription: 'How do you make this shift?' — prototype a flat R&D pod with high-agency people	Practical action — answers the implied 'what do I do Monday'
The weight of responsibility — employees must choose accountability	Caveat / second audience — shifts the ask from leaders to ICs
Agency > Agents reprise + Lovable's parented-agent model	Restatement with operational color — agents serve context distribution, not decision-making
Trust-based growth starts with trust-based teams (links to her prior 'Growth is a trust problem' post)	Extension — external trust requires internal trust as precondition

Key terms

Agency Agency
An employee's authorized capacity to act on context — having both the information needed to make a call AND the permission to make it without escalation
Agents Agentic AI
AI systems that 'wait to be told what to do' — useful for distributing context but structurally incapable of making the calls
Context
The information substrate required to make good decisions — currently hoarded behind manager titles and Slack channels
Parented agent Human as Orchestrator of Agents
An internal AI agent assigned a human 'parent' — the person with deepest context in that domain — who is responsible for keeping the agent accurate and current
Status theater AI Confidence Theater
Title-and-hierarchy signaling that exists to allocate permission and audibility rather than to coordinate work
Organizational muscle memory
Trained-in habit of waiting for permission; persists even after permissions are formally lifted
Trust-based growth / trust-based teams Trust
External customer trust is downstream of internal employee trust; you cannot 'build in public' or 'react quickly to users' if you don't let employees speak and decide

Propositions

explicitAI tools do not produce velocity if the surrounding org system is designed to prevent action.conf: high
supportsWhy Enterprise AI Has a Leadership Problem · p1 — Both locate the enterprise AI bottleneck in the organizational system and leadership rather than in the technology.
supportsThe Organization Is the Bottleneck · p1 — Both assert AI tooling yields no organizational velocity when the surrounding org system (trust, permission structures, delivery practice) prevents value from flowing.
supports9) Nils Brede Moe om AI-paradokset, autonomi, læringskultur og fremtidens team · p12 — Both locate AI's missing organizational returns in the surrounding org system rather than the tooling — Brede Moe blames missing joint practice change, Elena blames permission and trust structures.
explicitTitles function primarily as permission structures — they decide who gets context, who gets heard, who gets to decide.conf: high
explicitDecisions are slow because they are expensive, and expensive because they are slow — a self-reinforcing loop.conf: high
explicitThe cost of a wrong decision falls as decision velocity rises, because reversal becomes cheap.conf: high
explicitMiddle management is structurally a telephone-game between layers; its core function shrinks as context is distributed.conf: high
supportsAgents Over Bubbles · p4 — Agrees the middle-management coordination layer shrinks as agents distribute context, supporting Thompson's replacement thesis for that layer.
explicitcorroboratedAI agents distribute context but cannot replace high-agency employees because they lack agency.conf: high
refinesAgents Over Bubbles · p4 — Qualifies which humans agents can displace: coordination layers yes, but agents lack the agency to replace high-agency employees.
supports ←Presentation: Platform Engineering in the Age of AI · p15 — Independent corroboration, from a platform-engineering context, that AI agents distribute execution but cannot substitute for human agency in setting intent/goals.
explicitcontestedThe right way to begin the shift is to prototype a flat, high-autonomy pod rather than rewrite the whole org.conf: high
contradicts4) Morten Eikeland: Det finnes alltid et hierarki · p19 — Elena prescribes prototyping flat, high-autonomy pods; Eikeland holds leaderless self-managed teams are descriptively incoherent because coordination and single-spokesperson needs persist.
explicitExisting employees may not be able to absorb new autonomy because they've been trained out of using it.conf: high
extendsPresentation: The Human Scalability Problem: Why Your Teams Don't Scale Like Your Code · p5 — Adds a precondition for usable autonomy beyond trust and psychological safety: employees may have been trained out of exercising agency and cannot simply absorb it when granted.
explicitAutonomy carries personal accountability that many employees have never experienced and may not want.conf: high
explicitExternal customer trust cannot exist without prior internal employee trust.conf: high
implicitThe structural failure is leadership behavior (distrust signaling), not org design or compensation contracts.conf: high
implicitcontestedHigh-agency individuals are a stable trait you can select for, not a state you produce.conf: high
contradictsHow Fast-Growing Companies Can Make Better Decisions · p2 — Elena treats high agency as a stable trait you can select for; the structured-empowerment account holds full decentralization demands unrealistically qualified hires, so selection cannot staff autonomy at scale.
implicitThe reader of this piece is in a position to authorize a flat pilot pod.conf: high
implicitcontestedAnthropic's flat title structure is causally connected to its velocity (not, e.g., its hiring filter or capital).conf: medium
contradicts4) Morten Eikeland: Det finnes alltid et hierarki · p4 — Elena credits Anthropic's flat title structure with its velocity; Eikeland holds that suppressing visible hierarchy markers does not abolish hierarchy, it only makes it harder to read.

Arguments

Pushing AI tooling into a command-and-control org will not increase velocity. strength: high

Explicit premises:

AI requires context to produce good output
Employees in command-and-control orgs are denied context (title-gated access, boss-only decisions)
Without context, even the best tools produce bad or no decisions

Implicit premises:

The bottleneck on output is decision velocity, not execution capacity
AI tools enhance the marginal worker, not the marginal decision-maker

Inferences:

AI productivity studies that show null effects in large enterprises may be measuring trust deficits, not tool deficits
AI rollouts should be paired with permission-system audits

Middle management is structurally redundant in a high-context, high-agency org. strength: medium

Explicit premises:

Middle management primarily moves information and decisions up/down the chain
Cross-functional unblocking diminishes when teams are fewer and context is centralized
Career coaching can be democratized to non-manager roles

Implicit premises:

The coordination problems middle management currently solves are artifacts of the distrust structure, not intrinsic to scale
Information centralization (a paradox given the agency thesis) is feasible without re-creating gatekeeping

Inferences:

If 'centralize context, distribute decisions' is feasible, the operational question becomes: what mechanism distributes context at scale without recreating titles?

Leaders should prototype a flat pod with new people rather than retrofit existing teams. strength: medium

Explicit premises:

Whole-org rewrites fail
Existing employees have organizational muscle memory of waiting for permission
High-agency operators exist and can be hired in

Implicit premises:

Muscle memory is sticky enough that retraining is slower than rehiring
A pilot pod can demonstrate results that justify broader change

Inferences:

Existing managers are unlikely to be the right pilot members — this is a quiet displacement argument
'Innovation theater' risk: the pilot becomes a sandbox that never expands

Autonomy is a contract employees must explicitly opt into, not a gift leaders give. strength: high

Explicit premises:

Autonomy comes with accountability for failure
Many employees have not experienced this and may not want it
Compensation for accountability is the historical justification for leadership pay

Implicit premises:

Current comp structures will need to compensate ICs for accountability they newly carry
Some employees will rationally decline the upgrade

Inferences:

Pay equity discussions follow: if ICs carry decision accountability, the manager premium has to be re-justified or redistributed
Performance management systems calibrated for 'did you do what you were told' will mis-evaluate high-agency ICs

Narratives

When you get a manager promotion, you suddenly get added to a separate Slack channel with information you previously had no access to.
→ Shows — rather than argues — that titles function as permission structures. Makes the abstract claim concrete with a felt-recognition example any tech reader has lived.
Anthropic's visible structure is 'member of technical staff' or 'member of non-technical staff.'
→ Existence proof that a high-status, high-velocity AI-native org can run without title hierarchy — implicitly suggests title flattening is causally connected to AI-native velocity.
At Lovable, each internal AI agent has a human 'parent' responsible for keeping its context accurate.
→ Operational instantiation of the 'distribute context, don't distribute decision authority to AI' principle — shows agents as plumbing for human agency rather than substitutes for it.
Telephone-game image: middle management moving decisions and info 'up and down, down and up, over and over.'
→ Reframes middle management's primary function as low-value translation work — preparing the reader to accept that the role's structural need is shrinking.

Questions addressed

✅ Why are AI transformation initiatives failing to produce the productivity gains companies expect?
Because the bottleneck is permission, not capability — orgs deny employees the context and authority that would let new tools translate into velocity.
✅ Will AI agents replace employees?
No — agents lack agency; they amplify context distribution but cannot make the calls that high-agency humans make.
⭕ How should a leader actually begin the shift?
Hedged answer: prototype with a flat pod of high-agency people, possibly new hires. Elena explicitly disclaims certainty: 'I'm not sure how you switch it.'
✅ What is the operational test for whether your org actually trusts its employees?
Implicit test embedded in the piece: do junior employees have access to the same Slack channels and decision context as managers, and can they reverse a call same-day without escalation? If no, you don't trust them.
⭕ How do you distribute context at scale without recreating the gatekeeping you just removed?
Gestures at 'parented agents' as the mechanism but doesn't develop how this scales across a 1000+ person org.
⭕ If middle management is redundant, what happens to existing managers and to the comp ladder that depends on management titles?
Not addressed.
⭕ How is decision accountability allocated and compensated in a flat-agency org?
Acknowledges the tension ('that's why leaders get paid more: accountability') but doesn't propose a new comp or evaluation model.

Gaps

incompleteTreats trust as a leadership-behavior problem (leaders should give trust, employees should accept accountability) rather than an org-design or contract/incentive problem. There is no mechanism design — no discussion of decision rights documents, RACI alternatives, equity/comp restructuring, escalation policies, or measurement systems that would make agency stick when individual managers regress to control. Pattern-matches to the HBR Managers AI Productivity Boom piece: name a structural problem, prescribe individual remedies.
Read next: Reinertsen, 'The Principles of Product Development Flow' (decision rights and queue economics); Hamel & Zanini, 'Humanocracy' (post-bureaucratic org design with explicit mechanisms); Laloux, 'Reinventing Organizations' (advice process as a contract for distributed decisions); Spotify's matrix history and post-mortems; works on internal markets for decision rights.
incompleteThe Anthropic and Lovable examples are AI-native startups with hiring filters, capital, and mission selection that pre-select for high-agency operators. The piece does not address whether the model travels to a 5,000-person incumbent with regulatory exposure, union representation, or legacy systems.
Read next: Cutler 'AI Won't Save Your Kingdoms' (incumbents losing to challengers with different structures); case studies on Haier RenDanHeYi; Morning Star's self-management; ING's agile reorg post-mortems.
closed byWhat is the harder human-capital problem beneath token capital? — That extraction's open gap: the Anthropic/Lovable high-agency examples are AI-native startups that pre-select high-agency operators, and it doesn't address whether the model travels to a large incumbent. This essay supplies the structural obstacle — incumbents structurally select FOR org-chart optimizers and AGAINST high-agency frontier talent, so the model does NOT travel cleanly; the incumbent must first become 'porous' enough to hold uncorporate people with corporate-scale compute. (Partial closure: it names the obstacle without giving the mechanism to overcome it.)
incompleteThe accountability-and-comp argument is named but not operationalized. If ICs newly carry decision accountability, what is the manager premium for? She gestures at the tension and moves on. Without a comp/perf-review redesign, the agency upgrade is unilateral risk transfer.
Read next: Buurtzorg's flat-pay model; literature on profit-sharing and gainsharing; recent work on equity-as-accountability-comp in startups.
illogicalThere is a quiet tension between 'centralize information' (named as a benefit reducing the need for middle managers to coordinate) and 'distribute decisions' (the agency thesis). Centralized information typically implies central gatekeeping; distributed decisions imply distributed access. The piece doesn't say what mechanism resolves the tension at scale — 'parented agents' is gestured at but not specified.
Read next: Conway's Law literature; works on platform engineering as context distribution; Team Topologies (Skelton & Pais).
incompleteThe 'high-agency people' framing treats agency as a hireable trait rather than a structural product. This shifts responsibility back to selection rather than to the system. It also has uncomfortable selection-effect implications: who gets sorted out when 'high-agency' becomes the dominant hiring filter?
Read next: Edmondson 'The Fearless Organization' (psychological safety as the structural enabler of voice/agency); critiques of meritocratic narratives in tech hiring.
closed byWhat 60 Years of Data Reveals About How Men and Women Experience Leadership — That gap asks for the missing structural reframing (agency treated as a hireable individual trait rather than a structural product, and the selection effects of who gets 'sorted out') and, in its reading direction, for critiques of meritocratic narratives. This article supplies both: it reframes advancement-worthiness as a structural product of sponsorship, P&L access, and the broken rung (p8), and it delivers an empirical critique of meritocratic narratives — merit is interpreted not measured, and self-proclaimed meritocracies show more bias (p4). Partial closure: it addresses the structural-product and meritocracy legs but not the psychological-safety/Edmondson leg the gap also gestures toward.
incompleteNo engagement with the failure modes of the prescription: pilot pods that become innovation theater; flat pods that recreate informal hierarchies; high-velocity decision-making that produces irreversible decisions before reflection (e.g. customer trust violations). The closing trust-based-growth move actually heightens this risk — fast public action with low oversight.
Read next: Failure-mode analyses of holacracy at Zappos; post-mortems on flat orgs (Valve, GitHub pre-2014); Karen Ho 'Liquidated' on what happens when decision velocity outruns institutional memory.

Closes gaps elsewhere

Generative AI in the Real World: Aishwarya Naresh Reganti on Making AI Work in Production · g2 — Names the structural barriers the gap said were unaddressed: titles as permission structures, expensive-because-slow decision loops, and leadership distrust are why most orgs cannot run the tight cross-functional feedback loops AI products require (p1-p3, p11).
Presentation: The Human Scalability Problem: Why Your Teams Don't Scale Like Your Code · g2 — Explores what the talk dismissed: AI agents distribute context and shrink the middle-management translation problem, shifting the binding constraint of scaling from communication overload to trust and permission structures.

Your AI strategy has a trust problem, not a tooling problem

Structure

Key terms

Propositions

Arguments

Narratives

Questions addressed

Gaps

Closes gaps elsewhere

Related documents