I like Agile. I like self-discipline. I like programs that ship and programs
that study.
What I don’t like: tribes.
Within the final couple many years, many groups camped on the ends of a
spectrum:
- Conventional outlets handled optimization as advantage and adaptation as danger.
- Agile outlets handled adaptation as advantage and optimization as betrayal.
Each missed the purpose.
By adaptation I imply quick studying and course-correction below
uncertainty.
By optimization I imply reliability and repeatability below
constraints.
The error is treating both one as a everlasting working mode.
The grownup query is: what ought to dominate proper now?
This can be a pressure to handle, not a facet to select.
Why this issues now (past software program)
Software program groups have lived inside this pressure for years.
Now extra industries hit the identical wall—quick.
Life sciences (Biotech) offers clear examples. Instruments like CRISPR
(gene modifying), AlphaFold (3-D protein folding) and different AI-assisted
discovery fashions compress early cycles.
CRISPR‑primarily based instruments aided COVID‑19 analysis and goal discovery, whereas
platform applied sciences like mRNA and viral vectors have been the important thing enablers of
the one‑yr vaccine timeline. AlphaFold can typically do in hours on a
laptop what used to take months or years within the lab.
Utilizing these instruments groups can discover extra choices, quicker. That sounds
like pure upside—till you bear in mind the opposite facet of the stress:
downstream work will get costlier, extra constrained, and fewer
forgiving.
Quicker studying doesn’t take away constraints. It raises the price of
sloppy selections.
So the potential hole shifts. It’s not “Can we “do” Agile?” It’s:
Can we handle the Adaptation ↔ Optimization pressure on function—at
velocity?
What I imply by “two modes”
I exploit two modes as a sensible shorthand. They don’t seem to be philosophies.
They’re working patterns.
Discover mode (adaptation-dominant)
Function: scale back uncertainty quick.
Discover mode treats work as a collection of hypotheses.
- You run quick cycles: speculation → take a look at → sign → determination.
- You retain prices low so you may change course.
- You shield proof high quality sufficient to belief the sign.
Discover mode does not imply chaos. It means you optimize the
studying loop.
Exploit mode (optimization-dominant)
Function: scale back variance below constraints.
Exploit mode treats work as a system you need to run reliably.
- You tighten the method.
- You increase proof thresholds.
- You shield security, high quality, safety, traceability.
- You continue to adapt, however solely inside clear guardrails.
Exploit mode does not imply forms. It means you optimize
reliability.
One essential nuance: dominance, not purity
Each modes exist on a regular basis.
- Discover phases nonetheless want optimization (cycle time, proof hygiene, cease
guidelines). - Exploit phases nonetheless want adaptation (disciplined amendments, managed
experiments, risk-based exceptions).
Dominance retains you out of faith.
A bridge state: “Increase”
Utilizing the phrases discover and exploit typically brings to thoughts Kent Beck’s
discover–increase–extract. That connection is beneficial.
I see increase because the bridge state the place a promising sign strikes from
low cost studying to scaled proof.
In increase, you do three issues without delay:
- 1. Scale proof (extra instances, extra quantity, extra environments)
- 2. Elevate constraints (high quality, security, governance, integration
self-discipline) - 3. Cut back ambiguity (clear thresholds for the subsequent dedication)
Increase is the place many orgs pay the very best handoff tax, as a result of groups
maintain discover behaviors whereas the work now calls for exploit self-discipline.
The handoff tax
Most packages don’t fail inside a part.
They fail on the seams.
I name the hidden price at seams the handoff tax:
- translation failures (similar phrases, totally different that means)
- proof mismatch (totally different bars for “sufficient proof”)
- possession fog (too many votes, too many vetoes)
- traceability gaps (nobody can reconstruct why a alternative occurred)
If you need velocity, minimize handoff tax. It beats “doing Agile more durable.”
A fast detour: why bimodal IT backfired
One early “resolution” to this pressure was bimodal IT: put exploratory
work in a single lane and steady supply in one other—continuously as separate
organizational items.
On paper it seemed tidy.
In follow it changed into warring tribes. One facet turned the
innovation heroes. The opposite turned the steadiness police. Selections
bounced between them, handoff tax exploded, and leaders tried to handle
battle as a substitute of designing the work.
The lesson: you may’t outsource this pressure to an org chart. The
functionality has to reside in each one that makes selections—from staff
members to executives.
A concrete instance: Sciex and early integration
In 2004–2006 I labored with Sciex, an ISO-certified mass spectrometry
instrument agency. A crash in the course of a pattern run can smash an
experiment and waste irreplaceable samples.
After a yr plus working with software program groups we tackled a frightening
undertaking–growth of a brand new mass spec instrument.
We discovered the large killer to be integration debt (handoff tax)—the ache
you retailer up when {hardware}, firmware, and software program converge late.
ISO necessities stored governance actual. So we averted a false
alternative.
- Governance optimized for time, cash, and traceability.
- Execution tailored to uncertainty with quick suggestions loops and early
integration.
Then the Director of Product Growth pushed a easy shift:
- firmware delivered to {hardware} in iterations, paced by {hardware}’s take a look at
schedule - as soon as {hardware} reached “sufficient perform,” software program joined so as to add
functions—additionally in increments - they did not anticipate a completely populated digital board to begin
integration exams
Consequence:
- integration exams began sooner, so points surfaced earlier and resolved
quicker - integration stayed steady as soon as minimal {hardware} existed, so the standard
end-game useful resource spike disappeared - communication improved as a result of all teams participated in integration, not
simply on the panic stage
That’s dominance tuning within the wild:
- discover early the place uncertainty stays excessive
- increase as proof scales and constraints rise
- exploit as soon as reliability issues greater than possibility creation
Make dominance operational: 4 dials
If you need dominance with out debates, use dials.
- Uncertainty — what you have no idea but
- Danger — what breaks for those who guess incorrect
- Value of change — what a pivot prices in time, cash, credibility
- Proof threshold — how a lot proof you require earlier than you commit
Flip the dials, set dominance, then design the workflow to match.
Discover-dominant: tune the training loop
- quick cycle time from speculation → take a look at → sign → determination
- clear cease guidelines (kill weak bets quick)
- proof hygiene (assumptions, controls, reproducible notes)
Two widespread failures: gradual studying and messy proof.
Increase: scale proof and tighten constraints
- bigger samples, broader environments, extra integration factors
- rising governance self-discipline
- specific thresholds for the subsequent dedication
Two widespread failures: false certainty and late integration.
Exploit-dominant: adapt inside guardrails
- disciplined amendments, with triggers and clear rationale
- managed experiments (not unintentional variance)
- traceability you may defend below audit
Two widespread failures: compliance theater and hidden workarounds.
Determination rights: use DARE, not RACI
Velocity and accountability want clear determination rights. This isn’t
hierarchy worship.
Many orgs attain for RACI:Accountable, Accountable, Consulted,
Knowledgeable. In follow, RACI typically turns selections into calendar sludge
and well mannered vetoes.
Use DARE as a substitute: Deciders, Advisors, Recommenders, Execution
stakeholders.
DARE retains “servant management” and “self-organizing” (and their
cousins: “empowered groups,” “decentralized selections”) from sliding into
delicate anarchy: you can provide extra folks a voice with out giving everybody a
vote.
- Deciders: the one votes; typically one (however not completely)
- Advisors: sturdy voice, no veto
- Recommenders: construct choices and tradeoffs
- Execution stakeholders: execute the decision and floor constraints early
DARE works at each degree—from a product staff to the CEO workers—as a result of
the sample stays the identical:
- clear decider(s)
- actual enter
- actual choices
- quick dedication
DARE saves autonomy from turning into consensus-by-exhaustion.
Tailoring: deal with it as working design
Many groups deal with tailoring like weight reduction: begin with a giant technique,
minimize steps, hope velocity reveals up.
That’s disassembly.
Actual tailoring means design for match:
- maintain constraints that shield security, high quality, traceability
- maintain practices that shield studying velocity and possibility creation
- design seams so modes don’t battle one another
Tailoring additionally calls for judgment, and judgment stays scarce. You should buy
instruments and templates. You possibly can’t purchase discernment at scale.
The take-away
Cease promoting “Agile vs Conventional.” That story sells the issue.
Design for the stress:
- deal with discover, increase, exploit as a set of dominance patterns
- flip the dials on function
- minimize handoff tax at seams
- deal with tailoring as working design
The place do you pay the very best handoff tax at the moment—and which dial would
you flip first?